1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team, 2000-2004
7 * ---------------------------------------------------------------------------*/
10 #include "PosixSource.h"
13 /* Linux needs _GNU_SOURCE to get RTLD_DEFAULT from <dlfcn.h> and
14 MREMAP_MAYMOVE from <sys/mman.h>.
23 #include "sm/Storage.h"
26 #include "LinkerInternals.h"
29 #include "StgPrimFloat.h" // for __int_encodeFloat etc.
32 #if !defined(mingw32_HOST_OS)
33 #include "posix/Signals.h"
36 // get protos for is*()
39 #ifdef HAVE_SYS_TYPES_H
40 #include <sys/types.h>
48 #ifdef HAVE_SYS_STAT_H
52 #if defined(HAVE_DLFCN_H)
56 #if defined(cygwin32_HOST_OS)
61 #ifdef HAVE_SYS_TIME_H
65 #include <sys/fcntl.h>
66 #include <sys/termios.h>
67 #include <sys/utime.h>
68 #include <sys/utsname.h>
72 #if defined(linux_HOST_OS) || defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS) || defined(darwin_HOST_OS)
83 #if defined(linux_HOST_OS) || defined(solaris2_HOST_OS) || defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS)
84 # define OBJFORMAT_ELF
85 # include <regex.h> // regex is already used by dlopen() so this is OK
86 // to use here without requiring an additional lib
87 #elif defined(cygwin32_HOST_OS) || defined (mingw32_HOST_OS)
88 # define OBJFORMAT_PEi386
91 #elif defined(darwin_HOST_OS)
92 # define OBJFORMAT_MACHO
94 # include <mach-o/loader.h>
95 # include <mach-o/nlist.h>
96 # include <mach-o/reloc.h>
97 #if !defined(HAVE_DLFCN_H)
98 # include <mach-o/dyld.h>
100 #if defined(powerpc_HOST_ARCH)
101 # include <mach-o/ppc/reloc.h>
103 #if defined(x86_64_HOST_ARCH)
104 # include <mach-o/x86_64/reloc.h>
108 #if defined(x86_64_HOST_ARCH) && defined(darwin_HOST_OS)
112 /* Hash table mapping symbol names to Symbol */
113 static /*Str*/HashTable *symhash;
115 /* Hash table mapping symbol names to StgStablePtr */
116 static /*Str*/HashTable *stablehash;
118 /* List of currently loaded objects */
119 ObjectCode *objects = NULL; /* initially empty */
121 static HsInt loadOc( ObjectCode* oc );
122 static ObjectCode* mkOc( char *path, char *image, int imageSize,
123 char *archiveMemberName
125 #ifdef darwin_HOST_OS
131 #if defined(OBJFORMAT_ELF)
132 static int ocVerifyImage_ELF ( ObjectCode* oc );
133 static int ocGetNames_ELF ( ObjectCode* oc );
134 static int ocResolve_ELF ( ObjectCode* oc );
135 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
136 static int ocAllocateSymbolExtras_ELF ( ObjectCode* oc );
138 #elif defined(OBJFORMAT_PEi386)
139 static int ocVerifyImage_PEi386 ( ObjectCode* oc );
140 static int ocGetNames_PEi386 ( ObjectCode* oc );
141 static int ocResolve_PEi386 ( ObjectCode* oc );
142 static void *lookupSymbolInDLLs ( unsigned char *lbl );
143 static void zapTrailingAtSign ( unsigned char *sym );
144 #elif defined(OBJFORMAT_MACHO)
145 static int ocVerifyImage_MachO ( ObjectCode* oc );
146 static int ocGetNames_MachO ( ObjectCode* oc );
147 static int ocResolve_MachO ( ObjectCode* oc );
150 static int machoGetMisalignment( FILE * );
152 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
153 static int ocAllocateSymbolExtras_MachO ( ObjectCode* oc );
155 #ifdef powerpc_HOST_ARCH
156 static void machoInitSymbolsWithoutUnderscore( void );
160 /* on x86_64 we have a problem with relocating symbol references in
161 * code that was compiled without -fPIC. By default, the small memory
162 * model is used, which assumes that symbol references can fit in a
163 * 32-bit slot. The system dynamic linker makes this work for
164 * references to shared libraries by either (a) allocating a jump
165 * table slot for code references, or (b) moving the symbol at load
166 * time (and copying its contents, if necessary) for data references.
168 * We unfortunately can't tell whether symbol references are to code
169 * or data. So for now we assume they are code (the vast majority
170 * are), and allocate jump-table slots. Unfortunately this will
171 * SILENTLY generate crashing code for data references. This hack is
172 * enabled by X86_64_ELF_NONPIC_HACK.
174 * One workaround is to use shared Haskell libraries. This is
175 * coming. Another workaround is to keep the static libraries but
176 * compile them with -fPIC, because that will generate PIC references
177 * to data which can be relocated. The PIC code is still too green to
178 * do this systematically, though.
181 * See thread http://www.haskell.org/pipermail/cvs-ghc/2007-September/038458.html
183 * Naming Scheme for Symbol Macros
185 * SymI_*: symbol is internal to the RTS. It resides in an object
186 * file/library that is statically.
187 * SymE_*: symbol is external to the RTS library. It might be linked
190 * Sym*_HasProto : the symbol prototype is imported in an include file
191 * or defined explicitly
192 * Sym*_NeedsProto: the symbol is undefined and we add a dummy
193 * default proto extern void sym(void);
195 #define X86_64_ELF_NONPIC_HACK 1
197 /* Link objects into the lower 2Gb on x86_64. GHC assumes the
198 * small memory model on this architecture (see gcc docs,
201 * MAP_32BIT not available on OpenBSD/amd64
203 #if defined(x86_64_HOST_ARCH) && defined(MAP_32BIT)
204 #define TRY_MAP_32BIT MAP_32BIT
206 #define TRY_MAP_32BIT 0
210 * Due to the small memory model (see above), on x86_64 we have to map
211 * all our non-PIC object files into the low 2Gb of the address space
212 * (why 2Gb and not 4Gb? Because all addresses must be reachable
213 * using a 32-bit signed PC-relative offset). On Linux we can do this
214 * using the MAP_32BIT flag to mmap(), however on other OSs
215 * (e.g. *BSD, see #2063, and also on Linux inside Xen, see #2512), we
216 * can't do this. So on these systems, we have to pick a base address
217 * in the low 2Gb of the address space and try to allocate memory from
220 * We pick a default address based on the OS, but also make this
221 * configurable via an RTS flag (+RTS -xm)
223 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
225 #if defined(MAP_32BIT)
226 // Try to use MAP_32BIT
227 #define MMAP_32BIT_BASE_DEFAULT 0
230 #define MMAP_32BIT_BASE_DEFAULT 0x40000000
233 static void *mmap_32bit_base = (void *)MMAP_32BIT_BASE_DEFAULT;
236 /* MAP_ANONYMOUS is MAP_ANON on some systems, e.g. OpenBSD */
237 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
238 #define MAP_ANONYMOUS MAP_ANON
241 /* -----------------------------------------------------------------------------
242 * Built-in symbols from the RTS
245 typedef struct _RtsSymbolVal {
250 #define Maybe_Stable_Names SymI_HasProto(stg_mkWeakzh) \
251 SymI_HasProto(stg_mkWeakForeignEnvzh) \
252 SymI_HasProto(stg_makeStableNamezh) \
253 SymI_HasProto(stg_finalizzeWeakzh)
255 #if !defined (mingw32_HOST_OS)
256 #define RTS_POSIX_ONLY_SYMBOLS \
257 SymI_HasProto(__hscore_get_saved_termios) \
258 SymI_HasProto(__hscore_set_saved_termios) \
259 SymI_HasProto(shutdownHaskellAndSignal) \
260 SymI_HasProto(lockFile) \
261 SymI_HasProto(unlockFile) \
262 SymI_HasProto(signal_handlers) \
263 SymI_HasProto(stg_sig_install) \
264 SymI_NeedsProto(nocldstop)
267 #if defined (cygwin32_HOST_OS)
268 #define RTS_MINGW_ONLY_SYMBOLS /**/
269 /* Don't have the ability to read import libs / archives, so
270 * we have to stupidly list a lot of what libcygwin.a
273 #define RTS_CYGWIN_ONLY_SYMBOLS \
274 SymI_HasProto(regfree) \
275 SymI_HasProto(regexec) \
276 SymI_HasProto(regerror) \
277 SymI_HasProto(regcomp) \
278 SymI_HasProto(__errno) \
279 SymI_HasProto(access) \
280 SymI_HasProto(chmod) \
281 SymI_HasProto(chdir) \
282 SymI_HasProto(close) \
283 SymI_HasProto(creat) \
285 SymI_HasProto(dup2) \
286 SymI_HasProto(fstat) \
287 SymI_HasProto(fcntl) \
288 SymI_HasProto(getcwd) \
289 SymI_HasProto(getenv) \
290 SymI_HasProto(lseek) \
291 SymI_HasProto(open) \
292 SymI_HasProto(fpathconf) \
293 SymI_HasProto(pathconf) \
294 SymI_HasProto(stat) \
296 SymI_HasProto(tanh) \
297 SymI_HasProto(cosh) \
298 SymI_HasProto(sinh) \
299 SymI_HasProto(atan) \
300 SymI_HasProto(acos) \
301 SymI_HasProto(asin) \
307 SymI_HasProto(sqrt) \
308 SymI_HasProto(localtime_r) \
309 SymI_HasProto(gmtime_r) \
310 SymI_HasProto(mktime) \
311 SymI_NeedsProto(_imp___tzname) \
312 SymI_HasProto(gettimeofday) \
313 SymI_HasProto(timezone) \
314 SymI_HasProto(tcgetattr) \
315 SymI_HasProto(tcsetattr) \
316 SymI_HasProto(memcpy) \
317 SymI_HasProto(memmove) \
318 SymI_HasProto(realloc) \
319 SymI_HasProto(malloc) \
320 SymI_HasProto(free) \
321 SymI_HasProto(fork) \
322 SymI_HasProto(lstat) \
323 SymI_HasProto(isatty) \
324 SymI_HasProto(mkdir) \
325 SymI_HasProto(opendir) \
326 SymI_HasProto(readdir) \
327 SymI_HasProto(rewinddir) \
328 SymI_HasProto(closedir) \
329 SymI_HasProto(link) \
330 SymI_HasProto(mkfifo) \
331 SymI_HasProto(pipe) \
332 SymI_HasProto(read) \
333 SymI_HasProto(rename) \
334 SymI_HasProto(rmdir) \
335 SymI_HasProto(select) \
336 SymI_HasProto(system) \
337 SymI_HasProto(write) \
338 SymI_HasProto(strcmp) \
339 SymI_HasProto(strcpy) \
340 SymI_HasProto(strncpy) \
341 SymI_HasProto(strerror) \
342 SymI_HasProto(sigaddset) \
343 SymI_HasProto(sigemptyset) \
344 SymI_HasProto(sigprocmask) \
345 SymI_HasProto(umask) \
346 SymI_HasProto(uname) \
347 SymI_HasProto(unlink) \
348 SymI_HasProto(utime) \
349 SymI_HasProto(waitpid)
351 #elif !defined(mingw32_HOST_OS)
352 #define RTS_MINGW_ONLY_SYMBOLS /**/
353 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
354 #else /* defined(mingw32_HOST_OS) */
355 #define RTS_POSIX_ONLY_SYMBOLS /**/
356 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
358 #if HAVE_GETTIMEOFDAY
359 #define RTS_MINGW_GETTIMEOFDAY_SYM SymI_NeedsProto(gettimeofday)
361 #define RTS_MINGW_GETTIMEOFDAY_SYM /**/
364 #if HAVE___MINGW_VFPRINTF
365 #define RTS___MINGW_VFPRINTF_SYM SymI_HasProto(__mingw_vfprintf)
367 #define RTS___MINGW_VFPRINTF_SYM /**/
370 /* These are statically linked from the mingw libraries into the ghc
371 executable, so we have to employ this hack. */
372 #define RTS_MINGW_ONLY_SYMBOLS \
373 SymI_HasProto(stg_asyncReadzh) \
374 SymI_HasProto(stg_asyncWritezh) \
375 SymI_HasProto(stg_asyncDoProczh) \
376 SymI_HasProto(memset) \
377 SymI_HasProto(inet_ntoa) \
378 SymI_HasProto(inet_addr) \
379 SymI_HasProto(htonl) \
380 SymI_HasProto(recvfrom) \
381 SymI_HasProto(listen) \
382 SymI_HasProto(bind) \
383 SymI_HasProto(shutdown) \
384 SymI_HasProto(connect) \
385 SymI_HasProto(htons) \
386 SymI_HasProto(ntohs) \
387 SymI_HasProto(getservbyname) \
388 SymI_HasProto(getservbyport) \
389 SymI_HasProto(getprotobynumber) \
390 SymI_HasProto(getprotobyname) \
391 SymI_HasProto(gethostbyname) \
392 SymI_HasProto(gethostbyaddr) \
393 SymI_HasProto(gethostname) \
394 SymI_HasProto(strcpy) \
395 SymI_HasProto(strncpy) \
396 SymI_HasProto(abort) \
397 SymI_NeedsProto(_alloca) \
398 SymI_HasProto(isxdigit) \
399 SymI_HasProto(isupper) \
400 SymI_HasProto(ispunct) \
401 SymI_HasProto(islower) \
402 SymI_HasProto(isspace) \
403 SymI_HasProto(isprint) \
404 SymI_HasProto(isdigit) \
405 SymI_HasProto(iscntrl) \
406 SymI_HasProto(isalpha) \
407 SymI_HasProto(isalnum) \
408 SymI_HasProto(isascii) \
409 RTS___MINGW_VFPRINTF_SYM \
410 SymI_HasProto(strcmp) \
411 SymI_HasProto(memmove) \
412 SymI_HasProto(realloc) \
413 SymI_HasProto(malloc) \
415 SymI_HasProto(tanh) \
416 SymI_HasProto(cosh) \
417 SymI_HasProto(sinh) \
418 SymI_HasProto(atan) \
419 SymI_HasProto(acos) \
420 SymI_HasProto(asin) \
426 SymI_HasProto(sqrt) \
427 SymI_HasProto(powf) \
428 SymI_HasProto(tanhf) \
429 SymI_HasProto(coshf) \
430 SymI_HasProto(sinhf) \
431 SymI_HasProto(atanf) \
432 SymI_HasProto(acosf) \
433 SymI_HasProto(asinf) \
434 SymI_HasProto(tanf) \
435 SymI_HasProto(cosf) \
436 SymI_HasProto(sinf) \
437 SymI_HasProto(expf) \
438 SymI_HasProto(logf) \
439 SymI_HasProto(sqrtf) \
441 SymI_HasProto(erfc) \
442 SymI_HasProto(erff) \
443 SymI_HasProto(erfcf) \
444 SymI_HasProto(memcpy) \
445 SymI_HasProto(rts_InstallConsoleEvent) \
446 SymI_HasProto(rts_ConsoleHandlerDone) \
447 SymI_NeedsProto(mktime) \
448 SymI_NeedsProto(_imp___timezone) \
449 SymI_NeedsProto(_imp___tzname) \
450 SymI_NeedsProto(_imp__tzname) \
451 SymI_NeedsProto(_imp___iob) \
452 SymI_NeedsProto(_imp___osver) \
453 SymI_NeedsProto(localtime) \
454 SymI_NeedsProto(gmtime) \
455 SymI_NeedsProto(opendir) \
456 SymI_NeedsProto(readdir) \
457 SymI_NeedsProto(rewinddir) \
458 SymI_NeedsProto(_imp____mb_cur_max) \
459 SymI_NeedsProto(_imp___pctype) \
460 SymI_NeedsProto(__chkstk) \
461 RTS_MINGW_GETTIMEOFDAY_SYM \
462 SymI_NeedsProto(closedir)
466 #if defined(darwin_HOST_OS) && HAVE_PRINTF_LDBLSTUB
467 #define RTS_DARWIN_ONLY_SYMBOLS \
468 SymI_NeedsProto(asprintf$LDBLStub) \
469 SymI_NeedsProto(err$LDBLStub) \
470 SymI_NeedsProto(errc$LDBLStub) \
471 SymI_NeedsProto(errx$LDBLStub) \
472 SymI_NeedsProto(fprintf$LDBLStub) \
473 SymI_NeedsProto(fscanf$LDBLStub) \
474 SymI_NeedsProto(fwprintf$LDBLStub) \
475 SymI_NeedsProto(fwscanf$LDBLStub) \
476 SymI_NeedsProto(printf$LDBLStub) \
477 SymI_NeedsProto(scanf$LDBLStub) \
478 SymI_NeedsProto(snprintf$LDBLStub) \
479 SymI_NeedsProto(sprintf$LDBLStub) \
480 SymI_NeedsProto(sscanf$LDBLStub) \
481 SymI_NeedsProto(strtold$LDBLStub) \
482 SymI_NeedsProto(swprintf$LDBLStub) \
483 SymI_NeedsProto(swscanf$LDBLStub) \
484 SymI_NeedsProto(syslog$LDBLStub) \
485 SymI_NeedsProto(vasprintf$LDBLStub) \
486 SymI_NeedsProto(verr$LDBLStub) \
487 SymI_NeedsProto(verrc$LDBLStub) \
488 SymI_NeedsProto(verrx$LDBLStub) \
489 SymI_NeedsProto(vfprintf$LDBLStub) \
490 SymI_NeedsProto(vfscanf$LDBLStub) \
491 SymI_NeedsProto(vfwprintf$LDBLStub) \
492 SymI_NeedsProto(vfwscanf$LDBLStub) \
493 SymI_NeedsProto(vprintf$LDBLStub) \
494 SymI_NeedsProto(vscanf$LDBLStub) \
495 SymI_NeedsProto(vsnprintf$LDBLStub) \
496 SymI_NeedsProto(vsprintf$LDBLStub) \
497 SymI_NeedsProto(vsscanf$LDBLStub) \
498 SymI_NeedsProto(vswprintf$LDBLStub) \
499 SymI_NeedsProto(vswscanf$LDBLStub) \
500 SymI_NeedsProto(vsyslog$LDBLStub) \
501 SymI_NeedsProto(vwarn$LDBLStub) \
502 SymI_NeedsProto(vwarnc$LDBLStub) \
503 SymI_NeedsProto(vwarnx$LDBLStub) \
504 SymI_NeedsProto(vwprintf$LDBLStub) \
505 SymI_NeedsProto(vwscanf$LDBLStub) \
506 SymI_NeedsProto(warn$LDBLStub) \
507 SymI_NeedsProto(warnc$LDBLStub) \
508 SymI_NeedsProto(warnx$LDBLStub) \
509 SymI_NeedsProto(wcstold$LDBLStub) \
510 SymI_NeedsProto(wprintf$LDBLStub) \
511 SymI_NeedsProto(wscanf$LDBLStub)
513 #define RTS_DARWIN_ONLY_SYMBOLS
517 # define MAIN_CAP_SYM SymI_HasProto(MainCapability)
519 # define MAIN_CAP_SYM
522 #if !defined(mingw32_HOST_OS)
523 #define RTS_USER_SIGNALS_SYMBOLS \
524 SymI_HasProto(setIOManagerControlFd) \
525 SymI_HasProto(setIOManagerWakeupFd) \
526 SymI_HasProto(ioManagerWakeup) \
527 SymI_HasProto(blockUserSignals) \
528 SymI_HasProto(unblockUserSignals)
530 #define RTS_USER_SIGNALS_SYMBOLS \
531 SymI_HasProto(ioManagerWakeup) \
532 SymI_HasProto(sendIOManagerEvent) \
533 SymI_HasProto(readIOManagerEvent) \
534 SymI_HasProto(getIOManagerEvent) \
535 SymI_HasProto(console_handler)
538 #define RTS_LIBFFI_SYMBOLS \
539 SymE_NeedsProto(ffi_prep_cif) \
540 SymE_NeedsProto(ffi_call) \
541 SymE_NeedsProto(ffi_type_void) \
542 SymE_NeedsProto(ffi_type_float) \
543 SymE_NeedsProto(ffi_type_double) \
544 SymE_NeedsProto(ffi_type_sint64) \
545 SymE_NeedsProto(ffi_type_uint64) \
546 SymE_NeedsProto(ffi_type_sint32) \
547 SymE_NeedsProto(ffi_type_uint32) \
548 SymE_NeedsProto(ffi_type_sint16) \
549 SymE_NeedsProto(ffi_type_uint16) \
550 SymE_NeedsProto(ffi_type_sint8) \
551 SymE_NeedsProto(ffi_type_uint8) \
552 SymE_NeedsProto(ffi_type_pointer)
554 #ifdef TABLES_NEXT_TO_CODE
555 #define RTS_RET_SYMBOLS /* nothing */
557 #define RTS_RET_SYMBOLS \
558 SymI_HasProto(stg_enter_ret) \
559 SymI_HasProto(stg_gc_fun_ret) \
560 SymI_HasProto(stg_ap_v_ret) \
561 SymI_HasProto(stg_ap_f_ret) \
562 SymI_HasProto(stg_ap_d_ret) \
563 SymI_HasProto(stg_ap_l_ret) \
564 SymI_HasProto(stg_ap_n_ret) \
565 SymI_HasProto(stg_ap_p_ret) \
566 SymI_HasProto(stg_ap_pv_ret) \
567 SymI_HasProto(stg_ap_pp_ret) \
568 SymI_HasProto(stg_ap_ppv_ret) \
569 SymI_HasProto(stg_ap_ppp_ret) \
570 SymI_HasProto(stg_ap_pppv_ret) \
571 SymI_HasProto(stg_ap_pppp_ret) \
572 SymI_HasProto(stg_ap_ppppp_ret) \
573 SymI_HasProto(stg_ap_pppppp_ret)
576 /* Modules compiled with -ticky may mention ticky counters */
577 /* This list should marry up with the one in $(TOP)/includes/stg/Ticky.h */
578 #define RTS_TICKY_SYMBOLS \
579 SymI_NeedsProto(ticky_entry_ctrs) \
580 SymI_NeedsProto(top_ct) \
582 SymI_HasProto(ENT_VIA_NODE_ctr) \
583 SymI_HasProto(ENT_STATIC_THK_ctr) \
584 SymI_HasProto(ENT_DYN_THK_ctr) \
585 SymI_HasProto(ENT_STATIC_FUN_DIRECT_ctr) \
586 SymI_HasProto(ENT_DYN_FUN_DIRECT_ctr) \
587 SymI_HasProto(ENT_STATIC_CON_ctr) \
588 SymI_HasProto(ENT_DYN_CON_ctr) \
589 SymI_HasProto(ENT_STATIC_IND_ctr) \
590 SymI_HasProto(ENT_DYN_IND_ctr) \
591 SymI_HasProto(ENT_PERM_IND_ctr) \
592 SymI_HasProto(ENT_PAP_ctr) \
593 SymI_HasProto(ENT_AP_ctr) \
594 SymI_HasProto(ENT_AP_STACK_ctr) \
595 SymI_HasProto(ENT_BH_ctr) \
596 SymI_HasProto(UNKNOWN_CALL_ctr) \
597 SymI_HasProto(SLOW_CALL_v_ctr) \
598 SymI_HasProto(SLOW_CALL_f_ctr) \
599 SymI_HasProto(SLOW_CALL_d_ctr) \
600 SymI_HasProto(SLOW_CALL_l_ctr) \
601 SymI_HasProto(SLOW_CALL_n_ctr) \
602 SymI_HasProto(SLOW_CALL_p_ctr) \
603 SymI_HasProto(SLOW_CALL_pv_ctr) \
604 SymI_HasProto(SLOW_CALL_pp_ctr) \
605 SymI_HasProto(SLOW_CALL_ppv_ctr) \
606 SymI_HasProto(SLOW_CALL_ppp_ctr) \
607 SymI_HasProto(SLOW_CALL_pppv_ctr) \
608 SymI_HasProto(SLOW_CALL_pppp_ctr) \
609 SymI_HasProto(SLOW_CALL_ppppp_ctr) \
610 SymI_HasProto(SLOW_CALL_pppppp_ctr) \
611 SymI_HasProto(SLOW_CALL_OTHER_ctr) \
612 SymI_HasProto(ticky_slow_call_unevald) \
613 SymI_HasProto(SLOW_CALL_ctr) \
614 SymI_HasProto(MULTI_CHUNK_SLOW_CALL_ctr) \
615 SymI_HasProto(MULTI_CHUNK_SLOW_CALL_CHUNKS_ctr) \
616 SymI_HasProto(KNOWN_CALL_ctr) \
617 SymI_HasProto(KNOWN_CALL_TOO_FEW_ARGS_ctr) \
618 SymI_HasProto(KNOWN_CALL_EXTRA_ARGS_ctr) \
619 SymI_HasProto(SLOW_CALL_FUN_TOO_FEW_ctr) \
620 SymI_HasProto(SLOW_CALL_FUN_CORRECT_ctr) \
621 SymI_HasProto(SLOW_CALL_FUN_TOO_MANY_ctr) \
622 SymI_HasProto(SLOW_CALL_PAP_TOO_FEW_ctr) \
623 SymI_HasProto(SLOW_CALL_PAP_CORRECT_ctr) \
624 SymI_HasProto(SLOW_CALL_PAP_TOO_MANY_ctr) \
625 SymI_HasProto(SLOW_CALL_UNEVALD_ctr) \
626 SymI_HasProto(UPDF_OMITTED_ctr) \
627 SymI_HasProto(UPDF_PUSHED_ctr) \
628 SymI_HasProto(CATCHF_PUSHED_ctr) \
629 SymI_HasProto(UPDF_RCC_PUSHED_ctr) \
630 SymI_HasProto(UPDF_RCC_OMITTED_ctr) \
631 SymI_HasProto(UPD_SQUEEZED_ctr) \
632 SymI_HasProto(UPD_CON_IN_NEW_ctr) \
633 SymI_HasProto(UPD_CON_IN_PLACE_ctr) \
634 SymI_HasProto(UPD_PAP_IN_NEW_ctr) \
635 SymI_HasProto(UPD_PAP_IN_PLACE_ctr) \
636 SymI_HasProto(ALLOC_HEAP_ctr) \
637 SymI_HasProto(ALLOC_HEAP_tot) \
638 SymI_HasProto(ALLOC_FUN_ctr) \
639 SymI_HasProto(ALLOC_FUN_adm) \
640 SymI_HasProto(ALLOC_FUN_gds) \
641 SymI_HasProto(ALLOC_FUN_slp) \
642 SymI_HasProto(UPD_NEW_IND_ctr) \
643 SymI_HasProto(UPD_NEW_PERM_IND_ctr) \
644 SymI_HasProto(UPD_OLD_IND_ctr) \
645 SymI_HasProto(UPD_OLD_PERM_IND_ctr) \
646 SymI_HasProto(UPD_BH_UPDATABLE_ctr) \
647 SymI_HasProto(UPD_BH_SINGLE_ENTRY_ctr) \
648 SymI_HasProto(UPD_CAF_BH_UPDATABLE_ctr) \
649 SymI_HasProto(UPD_CAF_BH_SINGLE_ENTRY_ctr) \
650 SymI_HasProto(GC_SEL_ABANDONED_ctr) \
651 SymI_HasProto(GC_SEL_MINOR_ctr) \
652 SymI_HasProto(GC_SEL_MAJOR_ctr) \
653 SymI_HasProto(GC_FAILED_PROMOTION_ctr) \
654 SymI_HasProto(ALLOC_UP_THK_ctr) \
655 SymI_HasProto(ALLOC_SE_THK_ctr) \
656 SymI_HasProto(ALLOC_THK_adm) \
657 SymI_HasProto(ALLOC_THK_gds) \
658 SymI_HasProto(ALLOC_THK_slp) \
659 SymI_HasProto(ALLOC_CON_ctr) \
660 SymI_HasProto(ALLOC_CON_adm) \
661 SymI_HasProto(ALLOC_CON_gds) \
662 SymI_HasProto(ALLOC_CON_slp) \
663 SymI_HasProto(ALLOC_TUP_ctr) \
664 SymI_HasProto(ALLOC_TUP_adm) \
665 SymI_HasProto(ALLOC_TUP_gds) \
666 SymI_HasProto(ALLOC_TUP_slp) \
667 SymI_HasProto(ALLOC_BH_ctr) \
668 SymI_HasProto(ALLOC_BH_adm) \
669 SymI_HasProto(ALLOC_BH_gds) \
670 SymI_HasProto(ALLOC_BH_slp) \
671 SymI_HasProto(ALLOC_PRIM_ctr) \
672 SymI_HasProto(ALLOC_PRIM_adm) \
673 SymI_HasProto(ALLOC_PRIM_gds) \
674 SymI_HasProto(ALLOC_PRIM_slp) \
675 SymI_HasProto(ALLOC_PAP_ctr) \
676 SymI_HasProto(ALLOC_PAP_adm) \
677 SymI_HasProto(ALLOC_PAP_gds) \
678 SymI_HasProto(ALLOC_PAP_slp) \
679 SymI_HasProto(ALLOC_TSO_ctr) \
680 SymI_HasProto(ALLOC_TSO_adm) \
681 SymI_HasProto(ALLOC_TSO_gds) \
682 SymI_HasProto(ALLOC_TSO_slp) \
683 SymI_HasProto(RET_NEW_ctr) \
684 SymI_HasProto(RET_OLD_ctr) \
685 SymI_HasProto(RET_UNBOXED_TUP_ctr) \
686 SymI_HasProto(RET_SEMI_loads_avoided)
689 // On most platforms, the garbage collector rewrites references
690 // to small integer and char objects to a set of common, shared ones.
692 // We don't do this when compiling to Windows DLLs at the moment because
693 // it doesn't support cross package data references well.
695 #if defined(__PIC__) && defined(mingw32_HOST_OS)
696 #define RTS_INTCHAR_SYMBOLS
698 #define RTS_INTCHAR_SYMBOLS \
699 SymI_HasProto(stg_CHARLIKE_closure) \
700 SymI_HasProto(stg_INTLIKE_closure)
704 #define RTS_SYMBOLS \
707 SymI_HasProto(StgReturn) \
708 SymI_HasProto(stg_enter_info) \
709 SymI_HasProto(stg_gc_void_info) \
710 SymI_HasProto(__stg_gc_enter_1) \
711 SymI_HasProto(stg_gc_noregs) \
712 SymI_HasProto(stg_gc_unpt_r1_info) \
713 SymI_HasProto(stg_gc_unpt_r1) \
714 SymI_HasProto(stg_gc_unbx_r1_info) \
715 SymI_HasProto(stg_gc_unbx_r1) \
716 SymI_HasProto(stg_gc_f1_info) \
717 SymI_HasProto(stg_gc_f1) \
718 SymI_HasProto(stg_gc_d1_info) \
719 SymI_HasProto(stg_gc_d1) \
720 SymI_HasProto(stg_gc_l1_info) \
721 SymI_HasProto(stg_gc_l1) \
722 SymI_HasProto(__stg_gc_fun) \
723 SymI_HasProto(stg_gc_fun_info) \
724 SymI_HasProto(stg_gc_gen) \
725 SymI_HasProto(stg_gc_gen_info) \
726 SymI_HasProto(stg_gc_gen_hp) \
727 SymI_HasProto(stg_gc_ut) \
728 SymI_HasProto(stg_gen_yield) \
729 SymI_HasProto(stg_yield_noregs) \
730 SymI_HasProto(stg_yield_to_interpreter) \
731 SymI_HasProto(stg_gen_block) \
732 SymI_HasProto(stg_block_noregs) \
733 SymI_HasProto(stg_block_1) \
734 SymI_HasProto(stg_block_takemvar) \
735 SymI_HasProto(stg_block_putmvar) \
737 SymI_HasProto(MallocFailHook) \
738 SymI_HasProto(OnExitHook) \
739 SymI_HasProto(OutOfHeapHook) \
740 SymI_HasProto(StackOverflowHook) \
741 SymI_HasProto(addDLL) \
742 SymI_HasProto(__int_encodeDouble) \
743 SymI_HasProto(__word_encodeDouble) \
744 SymI_HasProto(__2Int_encodeDouble) \
745 SymI_HasProto(__int_encodeFloat) \
746 SymI_HasProto(__word_encodeFloat) \
747 SymI_HasProto(stg_atomicallyzh) \
748 SymI_HasProto(barf) \
749 SymI_HasProto(debugBelch) \
750 SymI_HasProto(errorBelch) \
751 SymI_HasProto(sysErrorBelch) \
752 SymI_HasProto(stg_getMaskingStatezh) \
753 SymI_HasProto(stg_maskAsyncExceptionszh) \
754 SymI_HasProto(stg_maskUninterruptiblezh) \
755 SymI_HasProto(stg_catchzh) \
756 SymI_HasProto(stg_catchRetryzh) \
757 SymI_HasProto(stg_catchSTMzh) \
758 SymI_HasProto(stg_checkzh) \
759 SymI_HasProto(closure_flags) \
760 SymI_HasProto(cmp_thread) \
761 SymI_HasProto(createAdjustor) \
762 SymI_HasProto(stg_decodeDoublezu2Intzh) \
763 SymI_HasProto(stg_decodeFloatzuIntzh) \
764 SymI_HasProto(defaultsHook) \
765 SymI_HasProto(stg_delayzh) \
766 SymI_HasProto(stg_deRefWeakzh) \
767 SymI_HasProto(stg_deRefStablePtrzh) \
768 SymI_HasProto(dirty_MUT_VAR) \
769 SymI_HasProto(stg_forkzh) \
770 SymI_HasProto(stg_forkOnzh) \
771 SymI_HasProto(forkProcess) \
772 SymI_HasProto(forkOS_createThread) \
773 SymI_HasProto(freeHaskellFunctionPtr) \
774 SymI_HasProto(getOrSetTypeableStore) \
775 SymI_HasProto(getOrSetGHCConcSignalSignalHandlerStore) \
776 SymI_HasProto(getOrSetGHCConcWindowsPendingDelaysStore) \
777 SymI_HasProto(getOrSetGHCConcWindowsIOManagerThreadStore) \
778 SymI_HasProto(getOrSetGHCConcWindowsProddingStore) \
779 SymI_HasProto(getOrSetSystemEventThreadEventManagerStore) \
780 SymI_HasProto(getOrSetSystemEventThreadIOManagerThreadStore) \
781 SymI_HasProto(genSymZh) \
782 SymI_HasProto(genericRaise) \
783 SymI_HasProto(getProgArgv) \
784 SymI_HasProto(getFullProgArgv) \
785 SymI_HasProto(getStablePtr) \
786 SymI_HasProto(hs_init) \
787 SymI_HasProto(hs_exit) \
788 SymI_HasProto(hs_set_argv) \
789 SymI_HasProto(hs_add_root) \
790 SymI_HasProto(hs_perform_gc) \
791 SymI_HasProto(hs_free_stable_ptr) \
792 SymI_HasProto(hs_free_fun_ptr) \
793 SymI_HasProto(hs_hpc_rootModule) \
794 SymI_HasProto(hs_hpc_module) \
795 SymI_HasProto(initLinker) \
796 SymI_HasProto(stg_unpackClosurezh) \
797 SymI_HasProto(stg_getApStackValzh) \
798 SymI_HasProto(stg_getSparkzh) \
799 SymI_HasProto(stg_numSparkszh) \
800 SymI_HasProto(stg_isCurrentThreadBoundzh) \
801 SymI_HasProto(stg_isEmptyMVarzh) \
802 SymI_HasProto(stg_killThreadzh) \
803 SymI_HasProto(loadArchive) \
804 SymI_HasProto(loadObj) \
805 SymI_HasProto(insertStableSymbol) \
806 SymI_HasProto(insertSymbol) \
807 SymI_HasProto(lookupSymbol) \
808 SymI_HasProto(stg_makeStablePtrzh) \
809 SymI_HasProto(stg_mkApUpd0zh) \
810 SymI_HasProto(stg_myThreadIdzh) \
811 SymI_HasProto(stg_labelThreadzh) \
812 SymI_HasProto(stg_newArrayzh) \
813 SymI_HasProto(stg_newBCOzh) \
814 SymI_HasProto(stg_newByteArrayzh) \
815 SymI_HasProto_redirect(newCAF, newDynCAF) \
816 SymI_HasProto(stg_newMVarzh) \
817 SymI_HasProto(stg_newMutVarzh) \
818 SymI_HasProto(stg_newTVarzh) \
819 SymI_HasProto(stg_noDuplicatezh) \
820 SymI_HasProto(stg_atomicModifyMutVarzh) \
821 SymI_HasProto(stg_newPinnedByteArrayzh) \
822 SymI_HasProto(stg_newAlignedPinnedByteArrayzh) \
823 SymI_HasProto(newSpark) \
824 SymI_HasProto(performGC) \
825 SymI_HasProto(performMajorGC) \
826 SymI_HasProto(prog_argc) \
827 SymI_HasProto(prog_argv) \
828 SymI_HasProto(stg_putMVarzh) \
829 SymI_HasProto(stg_raisezh) \
830 SymI_HasProto(stg_raiseIOzh) \
831 SymI_HasProto(stg_readTVarzh) \
832 SymI_HasProto(stg_readTVarIOzh) \
833 SymI_HasProto(resumeThread) \
834 SymI_HasProto(resolveObjs) \
835 SymI_HasProto(stg_retryzh) \
836 SymI_HasProto(rts_apply) \
837 SymI_HasProto(rts_checkSchedStatus) \
838 SymI_HasProto(rts_eval) \
839 SymI_HasProto(rts_evalIO) \
840 SymI_HasProto(rts_evalLazyIO) \
841 SymI_HasProto(rts_evalStableIO) \
842 SymI_HasProto(rts_eval_) \
843 SymI_HasProto(rts_getBool) \
844 SymI_HasProto(rts_getChar) \
845 SymI_HasProto(rts_getDouble) \
846 SymI_HasProto(rts_getFloat) \
847 SymI_HasProto(rts_getInt) \
848 SymI_HasProto(rts_getInt8) \
849 SymI_HasProto(rts_getInt16) \
850 SymI_HasProto(rts_getInt32) \
851 SymI_HasProto(rts_getInt64) \
852 SymI_HasProto(rts_getPtr) \
853 SymI_HasProto(rts_getFunPtr) \
854 SymI_HasProto(rts_getStablePtr) \
855 SymI_HasProto(rts_getThreadId) \
856 SymI_HasProto(rts_getWord) \
857 SymI_HasProto(rts_getWord8) \
858 SymI_HasProto(rts_getWord16) \
859 SymI_HasProto(rts_getWord32) \
860 SymI_HasProto(rts_getWord64) \
861 SymI_HasProto(rts_lock) \
862 SymI_HasProto(rts_mkBool) \
863 SymI_HasProto(rts_mkChar) \
864 SymI_HasProto(rts_mkDouble) \
865 SymI_HasProto(rts_mkFloat) \
866 SymI_HasProto(rts_mkInt) \
867 SymI_HasProto(rts_mkInt8) \
868 SymI_HasProto(rts_mkInt16) \
869 SymI_HasProto(rts_mkInt32) \
870 SymI_HasProto(rts_mkInt64) \
871 SymI_HasProto(rts_mkPtr) \
872 SymI_HasProto(rts_mkFunPtr) \
873 SymI_HasProto(rts_mkStablePtr) \
874 SymI_HasProto(rts_mkString) \
875 SymI_HasProto(rts_mkWord) \
876 SymI_HasProto(rts_mkWord8) \
877 SymI_HasProto(rts_mkWord16) \
878 SymI_HasProto(rts_mkWord32) \
879 SymI_HasProto(rts_mkWord64) \
880 SymI_HasProto(rts_unlock) \
881 SymI_HasProto(rts_unsafeGetMyCapability) \
882 SymI_HasProto(rtsSupportsBoundThreads) \
883 SymI_HasProto(rts_isProfiled) \
884 SymI_HasProto(setProgArgv) \
885 SymI_HasProto(startupHaskell) \
886 SymI_HasProto(shutdownHaskell) \
887 SymI_HasProto(shutdownHaskellAndExit) \
888 SymI_HasProto(stable_ptr_table) \
889 SymI_HasProto(stackOverflow) \
890 SymI_HasProto(stg_CAF_BLACKHOLE_info) \
891 SymI_HasProto(stg_BLACKHOLE_info) \
892 SymI_HasProto(__stg_EAGER_BLACKHOLE_info) \
893 SymI_HasProto(stg_BLOCKING_QUEUE_CLEAN_info) \
894 SymI_HasProto(stg_BLOCKING_QUEUE_DIRTY_info) \
895 SymI_HasProto(startTimer) \
896 SymI_HasProto(stg_MVAR_CLEAN_info) \
897 SymI_HasProto(stg_MVAR_DIRTY_info) \
898 SymI_HasProto(stg_IND_STATIC_info) \
899 SymI_HasProto(stg_ARR_WORDS_info) \
900 SymI_HasProto(stg_MUT_ARR_PTRS_DIRTY_info) \
901 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN_info) \
902 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN0_info) \
903 SymI_HasProto(stg_WEAK_info) \
904 SymI_HasProto(stg_ap_v_info) \
905 SymI_HasProto(stg_ap_f_info) \
906 SymI_HasProto(stg_ap_d_info) \
907 SymI_HasProto(stg_ap_l_info) \
908 SymI_HasProto(stg_ap_n_info) \
909 SymI_HasProto(stg_ap_p_info) \
910 SymI_HasProto(stg_ap_pv_info) \
911 SymI_HasProto(stg_ap_pp_info) \
912 SymI_HasProto(stg_ap_ppv_info) \
913 SymI_HasProto(stg_ap_ppp_info) \
914 SymI_HasProto(stg_ap_pppv_info) \
915 SymI_HasProto(stg_ap_pppp_info) \
916 SymI_HasProto(stg_ap_ppppp_info) \
917 SymI_HasProto(stg_ap_pppppp_info) \
918 SymI_HasProto(stg_ap_0_fast) \
919 SymI_HasProto(stg_ap_v_fast) \
920 SymI_HasProto(stg_ap_f_fast) \
921 SymI_HasProto(stg_ap_d_fast) \
922 SymI_HasProto(stg_ap_l_fast) \
923 SymI_HasProto(stg_ap_n_fast) \
924 SymI_HasProto(stg_ap_p_fast) \
925 SymI_HasProto(stg_ap_pv_fast) \
926 SymI_HasProto(stg_ap_pp_fast) \
927 SymI_HasProto(stg_ap_ppv_fast) \
928 SymI_HasProto(stg_ap_ppp_fast) \
929 SymI_HasProto(stg_ap_pppv_fast) \
930 SymI_HasProto(stg_ap_pppp_fast) \
931 SymI_HasProto(stg_ap_ppppp_fast) \
932 SymI_HasProto(stg_ap_pppppp_fast) \
933 SymI_HasProto(stg_ap_1_upd_info) \
934 SymI_HasProto(stg_ap_2_upd_info) \
935 SymI_HasProto(stg_ap_3_upd_info) \
936 SymI_HasProto(stg_ap_4_upd_info) \
937 SymI_HasProto(stg_ap_5_upd_info) \
938 SymI_HasProto(stg_ap_6_upd_info) \
939 SymI_HasProto(stg_ap_7_upd_info) \
940 SymI_HasProto(stg_exit) \
941 SymI_HasProto(stg_sel_0_upd_info) \
942 SymI_HasProto(stg_sel_10_upd_info) \
943 SymI_HasProto(stg_sel_11_upd_info) \
944 SymI_HasProto(stg_sel_12_upd_info) \
945 SymI_HasProto(stg_sel_13_upd_info) \
946 SymI_HasProto(stg_sel_14_upd_info) \
947 SymI_HasProto(stg_sel_15_upd_info) \
948 SymI_HasProto(stg_sel_1_upd_info) \
949 SymI_HasProto(stg_sel_2_upd_info) \
950 SymI_HasProto(stg_sel_3_upd_info) \
951 SymI_HasProto(stg_sel_4_upd_info) \
952 SymI_HasProto(stg_sel_5_upd_info) \
953 SymI_HasProto(stg_sel_6_upd_info) \
954 SymI_HasProto(stg_sel_7_upd_info) \
955 SymI_HasProto(stg_sel_8_upd_info) \
956 SymI_HasProto(stg_sel_9_upd_info) \
957 SymI_HasProto(stg_upd_frame_info) \
958 SymI_HasProto(stg_bh_upd_frame_info) \
959 SymI_HasProto(suspendThread) \
960 SymI_HasProto(stg_takeMVarzh) \
961 SymI_HasProto(stg_threadStatuszh) \
962 SymI_HasProto(stg_tryPutMVarzh) \
963 SymI_HasProto(stg_tryTakeMVarzh) \
964 SymI_HasProto(stg_unmaskAsyncExceptionszh) \
965 SymI_HasProto(unloadObj) \
966 SymI_HasProto(stg_unsafeThawArrayzh) \
967 SymI_HasProto(stg_waitReadzh) \
968 SymI_HasProto(stg_waitWritezh) \
969 SymI_HasProto(stg_writeTVarzh) \
970 SymI_HasProto(stg_yieldzh) \
971 SymI_NeedsProto(stg_interp_constr_entry) \
972 SymI_HasProto(stg_arg_bitmaps) \
973 SymI_HasProto(alloc_blocks_lim) \
975 SymI_HasProto(allocate) \
976 SymI_HasProto(allocateExec) \
977 SymI_HasProto(freeExec) \
978 SymI_HasProto(getAllocations) \
979 SymI_HasProto(revertCAFs) \
980 SymI_HasProto(RtsFlags) \
981 SymI_NeedsProto(rts_breakpoint_io_action) \
982 SymI_NeedsProto(rts_stop_next_breakpoint) \
983 SymI_NeedsProto(rts_stop_on_exception) \
984 SymI_HasProto(stopTimer) \
985 SymI_HasProto(n_capabilities) \
986 SymI_HasProto(stg_traceCcszh) \
987 SymI_HasProto(stg_traceEventzh) \
988 RTS_USER_SIGNALS_SYMBOLS \
992 // 64-bit support functions in libgcc.a
993 #if defined(__GNUC__) && SIZEOF_VOID_P <= 4
994 #define RTS_LIBGCC_SYMBOLS \
995 SymI_NeedsProto(__divdi3) \
996 SymI_NeedsProto(__udivdi3) \
997 SymI_NeedsProto(__moddi3) \
998 SymI_NeedsProto(__umoddi3) \
999 SymI_NeedsProto(__muldi3) \
1000 SymI_NeedsProto(__ashldi3) \
1001 SymI_NeedsProto(__ashrdi3) \
1002 SymI_NeedsProto(__lshrdi3)
1004 #define RTS_LIBGCC_SYMBOLS
1007 #if defined(darwin_HOST_OS) && defined(powerpc_HOST_ARCH)
1008 // Symbols that don't have a leading underscore
1009 // on Mac OS X. They have to receive special treatment,
1010 // see machoInitSymbolsWithoutUnderscore()
1011 #define RTS_MACHO_NOUNDERLINE_SYMBOLS \
1012 SymI_NeedsProto(saveFP) \
1013 SymI_NeedsProto(restFP)
1016 /* entirely bogus claims about types of these symbols */
1017 #define SymI_NeedsProto(vvv) extern void vvv(void);
1018 #if defined(__PIC__) && defined(mingw32_HOST_OS)
1019 #define SymE_HasProto(vvv) SymE_HasProto(vvv);
1020 #define SymE_NeedsProto(vvv) extern void _imp__ ## vvv (void);
1022 #define SymE_NeedsProto(vvv) SymI_NeedsProto(vvv);
1023 #define SymE_HasProto(vvv) SymI_HasProto(vvv)
1025 #define SymI_HasProto(vvv) /**/
1026 #define SymI_HasProto_redirect(vvv,xxx) /**/
1029 RTS_POSIX_ONLY_SYMBOLS
1030 RTS_MINGW_ONLY_SYMBOLS
1031 RTS_CYGWIN_ONLY_SYMBOLS
1032 RTS_DARWIN_ONLY_SYMBOLS
1035 #undef SymI_NeedsProto
1036 #undef SymI_HasProto
1037 #undef SymI_HasProto_redirect
1038 #undef SymE_HasProto
1039 #undef SymE_NeedsProto
1041 #ifdef LEADING_UNDERSCORE
1042 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
1044 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
1047 #define SymI_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1049 #define SymE_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1050 (void*)DLL_IMPORT_DATA_REF(vvv) },
1052 #define SymI_NeedsProto(vvv) SymI_HasProto(vvv)
1053 #define SymE_NeedsProto(vvv) SymE_HasProto(vvv)
1055 // SymI_HasProto_redirect allows us to redirect references to one symbol to
1056 // another symbol. See newCAF/newDynCAF for an example.
1057 #define SymI_HasProto_redirect(vvv,xxx) \
1058 { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1061 static RtsSymbolVal rtsSyms[] = {
1064 RTS_POSIX_ONLY_SYMBOLS
1065 RTS_MINGW_ONLY_SYMBOLS
1066 RTS_CYGWIN_ONLY_SYMBOLS
1067 RTS_DARWIN_ONLY_SYMBOLS
1070 #if defined(darwin_HOST_OS) && defined(i386_HOST_ARCH)
1071 // dyld stub code contains references to this,
1072 // but it should never be called because we treat
1073 // lazy pointers as nonlazy.
1074 { "dyld_stub_binding_helper", (void*)0xDEADBEEF },
1076 { 0, 0 } /* sentinel */
1081 /* -----------------------------------------------------------------------------
1082 * Insert symbols into hash tables, checking for duplicates.
1085 static void ghciInsertStrHashTable ( char* obj_name,
1091 if (lookupHashTable(table, (StgWord)key) == NULL)
1093 insertStrHashTable(table, (StgWord)key, data);
1098 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
1100 "whilst processing object file\n"
1102 "This could be caused by:\n"
1103 " * Loading two different object files which export the same symbol\n"
1104 " * Specifying the same object file twice on the GHCi command line\n"
1105 " * An incorrect `package.conf' entry, causing some object to be\n"
1107 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
1114 /* -----------------------------------------------------------------------------
1115 * initialize the object linker
1119 static int linker_init_done = 0 ;
1121 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1122 static void *dl_prog_handle;
1123 static regex_t re_invalid;
1124 static regex_t re_realso;
1126 static Mutex dl_mutex; // mutex to protect dlopen/dlerror critical section
1134 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1138 IF_DEBUG(linker, debugBelch("initLinker: start\n"));
1140 /* Make initLinker idempotent, so we can call it
1141 before evey relevant operation; that means we
1142 don't need to initialise the linker separately */
1143 if (linker_init_done == 1) {
1144 IF_DEBUG(linker, debugBelch("initLinker: idempotent return\n"));
1147 linker_init_done = 1;
1150 #if defined(THREADED_RTS) && (defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO))
1151 initMutex(&dl_mutex);
1153 stablehash = allocStrHashTable();
1154 symhash = allocStrHashTable();
1156 /* populate the symbol table with stuff from the RTS */
1157 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
1158 ghciInsertStrHashTable("(GHCi built-in symbols)",
1159 symhash, sym->lbl, sym->addr);
1160 IF_DEBUG(linker, debugBelch("initLinker: inserting rts symbol %s, %p\n", sym->lbl, sym->addr));
1162 # if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH)
1163 machoInitSymbolsWithoutUnderscore();
1166 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1167 # if defined(RTLD_DEFAULT)
1168 dl_prog_handle = RTLD_DEFAULT;
1170 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
1171 # endif /* RTLD_DEFAULT */
1173 compileResult = regcomp(&re_invalid,
1174 "(([^ \t()])+\\.so([^ \t:()])*):([ \t])*invalid ELF header",
1176 ASSERT( compileResult == 0 );
1177 compileResult = regcomp(&re_realso,
1178 "GROUP *\\( *(([^ )])+)",
1180 ASSERT( compileResult == 0 );
1183 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
1184 if (RtsFlags.MiscFlags.linkerMemBase != 0) {
1185 // User-override for mmap_32bit_base
1186 mmap_32bit_base = (void*)RtsFlags.MiscFlags.linkerMemBase;
1190 #if defined(mingw32_HOST_OS)
1192 * These two libraries cause problems when added to the static link,
1193 * but are necessary for resolving symbols in GHCi, hence we load
1194 * them manually here.
1200 IF_DEBUG(linker, debugBelch("initLinker: done\n"));
1205 exitLinker( void ) {
1206 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1207 if (linker_init_done == 1) {
1208 regfree(&re_invalid);
1209 regfree(&re_realso);
1211 closeMutex(&dl_mutex);
1217 /* -----------------------------------------------------------------------------
1218 * Loading DLL or .so dynamic libraries
1219 * -----------------------------------------------------------------------------
1221 * Add a DLL from which symbols may be found. In the ELF case, just
1222 * do RTLD_GLOBAL-style add, so no further messing around needs to
1223 * happen in order that symbols in the loaded .so are findable --
1224 * lookupSymbol() will subsequently see them by dlsym on the program's
1225 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
1227 * In the PEi386 case, open the DLLs and put handles to them in a
1228 * linked list. When looking for a symbol, try all handles in the
1229 * list. This means that we need to load even DLLs that are guaranteed
1230 * to be in the ghc.exe image already, just so we can get a handle
1231 * to give to loadSymbol, so that we can find the symbols. For such
1232 * libraries, the LoadLibrary call should be a no-op except for returning
1237 #if defined(OBJFORMAT_PEi386)
1238 /* A record for storing handles into DLLs. */
1243 struct _OpenedDLL* next;
1248 /* A list thereof. */
1249 static OpenedDLL* opened_dlls = NULL;
1252 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1255 internal_dlopen(const char *dll_name)
1261 // omitted: RTLD_NOW
1262 // see http://www.haskell.org/pipermail/cvs-ghc/2007-September/038570.html
1264 debugBelch("internal_dlopen: dll_name = '%s'\n", dll_name));
1266 //-------------- Begin critical section ------------------
1267 // This critical section is necessary because dlerror() is not
1268 // required to be reentrant (see POSIX -- IEEE Std 1003.1-2008)
1269 // Also, the error message returned must be copied to preserve it
1272 ACQUIRE_LOCK(&dl_mutex);
1273 hdl = dlopen(dll_name, RTLD_LAZY | RTLD_GLOBAL);
1277 /* dlopen failed; return a ptr to the error msg. */
1279 if (errmsg == NULL) errmsg = "addDLL: unknown error";
1280 errmsg_copy = stgMallocBytes(strlen(errmsg)+1, "addDLL");
1281 strcpy(errmsg_copy, errmsg);
1282 errmsg = errmsg_copy;
1284 RELEASE_LOCK(&dl_mutex);
1285 //--------------- End critical section -------------------
1292 addDLL( char *dll_name )
1294 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1295 /* ------------------- ELF DLL loader ------------------- */
1298 regmatch_t match[NMATCH];
1301 size_t match_length;
1302 #define MAXLINE 1000
1308 IF_DEBUG(linker, debugBelch("addDLL: dll_name = '%s'\n", dll_name));
1309 errmsg = internal_dlopen(dll_name);
1311 if (errmsg == NULL) {
1315 // GHC Trac ticket #2615
1316 // On some systems (e.g., Gentoo Linux) dynamic files (e.g. libc.so)
1317 // contain linker scripts rather than ELF-format object code. This
1318 // code handles the situation by recognizing the real object code
1319 // file name given in the linker script.
1321 // If an "invalid ELF header" error occurs, it is assumed that the
1322 // .so file contains a linker script instead of ELF object code.
1323 // In this case, the code looks for the GROUP ( ... ) linker
1324 // directive. If one is found, the first file name inside the
1325 // parentheses is treated as the name of a dynamic library and the
1326 // code attempts to dlopen that file. If this is also unsuccessful,
1327 // an error message is returned.
1329 // see if the error message is due to an invalid ELF header
1330 IF_DEBUG(linker, debugBelch("errmsg = '%s'\n", errmsg));
1331 result = regexec(&re_invalid, errmsg, (size_t) NMATCH, match, 0);
1332 IF_DEBUG(linker, debugBelch("result = %i\n", result));
1334 // success -- try to read the named file as a linker script
1335 match_length = (size_t) stg_min((match[1].rm_eo - match[1].rm_so),
1337 strncpy(line, (errmsg+(match[1].rm_so)),match_length);
1338 line[match_length] = '\0'; // make sure string is null-terminated
1339 IF_DEBUG(linker, debugBelch ("file name = '%s'\n", line));
1340 if ((fp = fopen(line, "r")) == NULL) {
1341 return errmsg; // return original error if open fails
1343 // try to find a GROUP ( ... ) command
1344 while (fgets(line, MAXLINE, fp) != NULL) {
1345 IF_DEBUG(linker, debugBelch("input line = %s", line));
1346 if (regexec(&re_realso, line, (size_t) NMATCH, match, 0) == 0) {
1347 // success -- try to dlopen the first named file
1348 IF_DEBUG(linker, debugBelch("match%s\n",""));
1349 line[match[1].rm_eo] = '\0';
1350 errmsg = internal_dlopen(line+match[1].rm_so);
1353 // if control reaches here, no GROUP ( ... ) directive was found
1354 // and the original error message is returned to the caller
1360 # elif defined(OBJFORMAT_PEi386)
1361 /* ------------------- Win32 DLL loader ------------------- */
1369 /* debugBelch("\naddDLL; dll_name = `%s'\n", dll_name); */
1371 /* See if we've already got it, and ignore if so. */
1372 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
1373 if (0 == strcmp(o_dll->name, dll_name))
1377 /* The file name has no suffix (yet) so that we can try
1378 both foo.dll and foo.drv
1380 The documentation for LoadLibrary says:
1381 If no file name extension is specified in the lpFileName
1382 parameter, the default library extension .dll is
1383 appended. However, the file name string can include a trailing
1384 point character (.) to indicate that the module name has no
1387 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
1388 sprintf(buf, "%s.DLL", dll_name);
1389 instance = LoadLibrary(buf);
1390 if (instance == NULL) {
1391 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1392 // KAA: allow loading of drivers (like winspool.drv)
1393 sprintf(buf, "%s.DRV", dll_name);
1394 instance = LoadLibrary(buf);
1395 if (instance == NULL) {
1396 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1397 // #1883: allow loading of unix-style libfoo.dll DLLs
1398 sprintf(buf, "lib%s.DLL", dll_name);
1399 instance = LoadLibrary(buf);
1400 if (instance == NULL) {
1407 /* Add this DLL to the list of DLLs in which to search for symbols. */
1408 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
1409 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
1410 strcpy(o_dll->name, dll_name);
1411 o_dll->instance = instance;
1412 o_dll->next = opened_dlls;
1413 opened_dlls = o_dll;
1419 sysErrorBelch(dll_name);
1421 /* LoadLibrary failed; return a ptr to the error msg. */
1422 return "addDLL: could not load DLL";
1425 barf("addDLL: not implemented on this platform");
1429 /* -----------------------------------------------------------------------------
1430 * insert a stable symbol in the hash table
1434 insertStableSymbol(char* obj_name, char* key, StgPtr p)
1436 ghciInsertStrHashTable(obj_name, stablehash, key, getStablePtr(p));
1440 /* -----------------------------------------------------------------------------
1441 * insert a symbol in the hash table
1444 insertSymbol(char* obj_name, char* key, void* data)
1446 ghciInsertStrHashTable(obj_name, symhash, key, data);
1449 /* -----------------------------------------------------------------------------
1450 * lookup a symbol in the hash table
1453 lookupSymbol( char *lbl )
1456 IF_DEBUG(linker, debugBelch("lookupSymbol: looking up %s\n", lbl));
1458 ASSERT(symhash != NULL);
1459 val = lookupStrHashTable(symhash, lbl);
1462 IF_DEBUG(linker, debugBelch("lookupSymbol: symbol not found\n"));
1463 # if defined(OBJFORMAT_ELF)
1464 return dlsym(dl_prog_handle, lbl);
1465 # elif defined(OBJFORMAT_MACHO)
1467 /* On OS X 10.3 and later, we use dlsym instead of the old legacy
1470 HACK: On OS X, global symbols are prefixed with an underscore.
1471 However, dlsym wants us to omit the leading underscore from the
1472 symbol name. For now, we simply strip it off here (and ONLY
1475 IF_DEBUG(linker, debugBelch("lookupSymbol: looking up %s with dlsym\n", lbl));
1476 ASSERT(lbl[0] == '_');
1477 return dlsym(dl_prog_handle, lbl+1);
1479 if(NSIsSymbolNameDefined(lbl)) {
1480 NSSymbol symbol = NSLookupAndBindSymbol(lbl);
1481 return NSAddressOfSymbol(symbol);
1485 # endif /* HAVE_DLFCN_H */
1486 # elif defined(OBJFORMAT_PEi386)
1489 sym = lookupSymbolInDLLs((unsigned char*)lbl);
1490 if (sym != NULL) { return sym; };
1492 // Also try looking up the symbol without the @N suffix. Some
1493 // DLLs have the suffixes on their symbols, some don't.
1494 zapTrailingAtSign ( (unsigned char*)lbl );
1495 sym = lookupSymbolInDLLs((unsigned char*)lbl);
1496 if (sym != NULL) { return sym; };
1504 IF_DEBUG(linker, debugBelch("lookupSymbol: value of %s is %p\n", lbl, val));
1509 /* -----------------------------------------------------------------------------
1510 * Debugging aid: look in GHCi's object symbol tables for symbols
1511 * within DELTA bytes of the specified address, and show their names.
1514 void ghci_enquire ( char* addr );
1516 void ghci_enquire ( char* addr )
1521 const int DELTA = 64;
1526 for (oc = objects; oc; oc = oc->next) {
1527 for (i = 0; i < oc->n_symbols; i++) {
1528 sym = oc->symbols[i];
1529 if (sym == NULL) continue;
1532 a = lookupStrHashTable(symhash, sym);
1535 // debugBelch("ghci_enquire: can't find %s\n", sym);
1537 else if (addr-DELTA <= a && a <= addr+DELTA) {
1538 debugBelch("%p + %3d == `%s'\n", addr, (int)(a - addr), sym);
1546 #define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))
1549 mmapForLinker (size_t bytes, nat flags, int fd)
1551 void *map_addr = NULL;
1554 static nat fixed = 0;
1556 pagesize = getpagesize();
1557 size = ROUND_UP(bytes, pagesize);
1559 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
1562 if (mmap_32bit_base != 0) {
1563 map_addr = mmap_32bit_base;
1567 result = mmap(map_addr, size, PROT_EXEC|PROT_READ|PROT_WRITE,
1568 MAP_PRIVATE|TRY_MAP_32BIT|fixed|flags, fd, 0);
1570 if (result == MAP_FAILED) {
1571 sysErrorBelch("mmap %lu bytes at %p",(lnat)size,map_addr);
1572 errorBelch("Try specifying an address with +RTS -xm<addr> -RTS");
1573 stg_exit(EXIT_FAILURE);
1576 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
1577 if (mmap_32bit_base != 0) {
1578 if (result == map_addr) {
1579 mmap_32bit_base = (StgWord8*)map_addr + size;
1581 if ((W_)result > 0x80000000) {
1582 // oops, we were given memory over 2Gb
1583 #if defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS)
1584 // Some platforms require MAP_FIXED. This is normally
1585 // a bad idea, because MAP_FIXED will overwrite
1586 // existing mappings.
1587 munmap(result,size);
1591 barf("loadObj: failed to mmap() memory below 2Gb; asked for %lu bytes at %p. Try specifying an address with +RTS -xm<addr> -RTS", size, map_addr, result);
1594 // hmm, we were given memory somewhere else, but it's
1595 // still under 2Gb so we can use it. Next time, ask
1596 // for memory right after the place we just got some
1597 mmap_32bit_base = (StgWord8*)result + size;
1601 if ((W_)result > 0x80000000) {
1602 // oops, we were given memory over 2Gb
1603 // ... try allocating memory somewhere else?;
1604 debugTrace(DEBUG_linker,"MAP_32BIT didn't work; gave us %lu bytes at 0x%p", bytes, result);
1605 munmap(result, size);
1607 // Set a base address and try again... (guess: 1Gb)
1608 mmap_32bit_base = (void*)0x40000000;
1619 mkOc( char *path, char *image, int imageSize,
1620 char *archiveMemberName
1622 #ifdef darwin_HOST_OS
1629 oc = stgMallocBytes(sizeof(ObjectCode), "loadArchive(oc)");
1631 # if defined(OBJFORMAT_ELF)
1632 oc->formatName = "ELF";
1633 # elif defined(OBJFORMAT_PEi386)
1634 oc->formatName = "PEi386";
1635 # elif defined(OBJFORMAT_MACHO)
1636 oc->formatName = "Mach-O";
1639 barf("loadObj: not implemented on this platform");
1643 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1644 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1645 strcpy(oc->fileName, path);
1647 if (archiveMemberName) {
1648 oc->archiveMemberName = stgMallocBytes( strlen(archiveMemberName)+1, "loadObj" );
1649 strcpy(oc->archiveMemberName, archiveMemberName);
1652 oc->archiveMemberName = NULL;
1655 oc->fileSize = imageSize;
1657 oc->sections = NULL;
1658 oc->proddables = NULL;
1661 #ifdef darwin_HOST_OS
1662 oc->misalignment = misalignment;
1666 /* chain it onto the list of objects */
1674 loadArchive( char *path )
1681 size_t fileNameSize;
1687 IF_DEBUG(linker, debugBelch("loadArchive: Loading archive `%s'\n", path));
1690 file = stgMallocBytes(fileSize, "loadArchive(file)");
1692 f = fopen(path, "rb");
1694 barf("loadObj: can't read `%s'", path);
1696 n = fread ( tmp, 1, 8, f );
1697 if (strncmp(tmp, "!<arch>\n", 8) != 0)
1698 barf("loadArchive: Not an archive: `%s'", path);
1701 n = fread ( file, 1, 16, f );
1707 barf("loadArchive: Failed reading file name from `%s'", path);
1710 n = fread ( tmp, 1, 12, f );
1712 barf("loadArchive: Failed reading mod time from `%s'", path);
1713 n = fread ( tmp, 1, 6, f );
1715 barf("loadArchive: Failed reading owner from `%s'", path);
1716 n = fread ( tmp, 1, 6, f );
1718 barf("loadArchive: Failed reading group from `%s'", path);
1719 n = fread ( tmp, 1, 8, f );
1721 barf("loadArchive: Failed reading mode from `%s'", path);
1722 n = fread ( tmp, 1, 10, f );
1724 barf("loadArchive: Failed reading size from `%s'", path);
1726 for (n = 0; isdigit(tmp[n]); n++);
1728 imageSize = atoi(tmp);
1729 n = fread ( tmp, 1, 2, f );
1730 if (strncmp(tmp, "\x60\x0A", 2) != 0)
1731 barf("loadArchive: Failed reading magic from `%s' at %ld. Got %c%c",
1732 path, ftell(f), tmp[0], tmp[1]);
1734 /* Check for BSD-variant large filenames */
1735 if (0 == strncmp(file, "#1/", 3)) {
1737 for (n = 3; isdigit(file[n]); n++);
1739 fileNameSize = atoi(file + 3);
1740 imageSize -= fileNameSize;
1741 if (fileNameSize > fileSize) {
1742 /* Double it to avoid potentially continually
1743 increasing it by 1 */
1744 fileSize = fileNameSize * 2;
1745 file = stgReallocBytes(file, fileSize, "loadArchive(file)");
1747 n = fread ( file, 1, fileNameSize, f );
1748 if (n != (int)fileNameSize)
1749 barf("loadArchive: Failed reading filename from `%s'", path);
1756 debugBelch("loadArchive: Found member file `%s'\n", file));
1759 for (n = 0; n < (int)fileNameSize - 1; n++) {
1760 if ((file[n] == '.') && (file[n + 1] == 'o')) {
1767 char *archiveMemberName;
1769 IF_DEBUG(linker, debugBelch("loadArchive: Member is an object file...loading...\n"));
1771 /* We can't mmap from the archive directly, as object
1772 files need to be 8-byte aligned but files in .ar
1773 archives are 2-byte aligned. When possible we use mmap
1774 to get some anonymous memory, as on 64-bit platforms if
1775 we use malloc then we can be given memory above 2^32.
1776 In the mmap case we're probably wasting lots of space;
1777 we could do better. */
1779 image = mmapForLinker(imageSize, MAP_ANONYMOUS, -1);
1781 image = stgMallocBytes(imageSize, "loadArchive(image)");
1783 n = fread ( image, 1, imageSize, f );
1785 barf("loadObj: error whilst reading `%s'", path);
1787 archiveMemberName = stgMallocBytes(strlen(path) + fileNameSize + 3,
1788 "loadArchive(file)");
1789 sprintf(archiveMemberName, "%s(%.*s)",
1790 path, (int)fileNameSize, file);
1792 oc = mkOc(path, image, imageSize, archiveMemberName
1794 #ifdef darwin_HOST_OS
1800 stgFree(archiveMemberName);
1802 if (0 == loadOc(oc)) {
1808 n = fseek(f, imageSize, SEEK_CUR);
1810 barf("loadArchive: error whilst seeking by %d in `%s'",
1813 /* .ar files are 2-byte aligned */
1814 if (imageSize % 2) {
1815 n = fread ( tmp, 1, 1, f );
1821 barf("loadArchive: Failed reading padding from `%s'", path);
1833 /* -----------------------------------------------------------------------------
1834 * Load an obj (populate the global symbol table, but don't resolve yet)
1836 * Returns: 1 if ok, 0 on error.
1839 loadObj( char *path )
1851 IF_DEBUG(linker, debugBelch("loadObj %s\n", path));
1855 /* debugBelch("loadObj %s\n", path ); */
1857 /* Check that we haven't already loaded this object.
1858 Ignore requests to load multiple times */
1862 for (o = objects; o; o = o->next) {
1863 if (0 == strcmp(o->fileName, path)) {
1865 break; /* don't need to search further */
1869 IF_DEBUG(linker, debugBelch(
1870 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1871 "same object file twice:\n"
1873 "GHCi will ignore this, but be warned.\n"
1875 return 1; /* success */
1879 r = stat(path, &st);
1881 IF_DEBUG(linker, debugBelch("File doesn't exist\n"));
1885 fileSize = st.st_size;
1888 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1890 #if defined(openbsd_HOST_OS)
1891 fd = open(path, O_RDONLY, S_IRUSR);
1893 fd = open(path, O_RDONLY);
1896 barf("loadObj: can't open `%s'", path);
1898 image = mmapForLinker(fileSize, 0, fd);
1902 #else /* !USE_MMAP */
1903 /* load the image into memory */
1904 f = fopen(path, "rb");
1906 barf("loadObj: can't read `%s'", path);
1908 # if defined(mingw32_HOST_OS)
1909 // TODO: We would like to use allocateExec here, but allocateExec
1910 // cannot currently allocate blocks large enough.
1911 image = VirtualAlloc(NULL, fileSize, MEM_RESERVE | MEM_COMMIT,
1912 PAGE_EXECUTE_READWRITE);
1913 # elif defined(darwin_HOST_OS)
1914 // In a Mach-O .o file, all sections can and will be misaligned
1915 // if the total size of the headers is not a multiple of the
1916 // desired alignment. This is fine for .o files that only serve
1917 // as input for the static linker, but it's not fine for us,
1918 // as SSE (used by gcc for floating point) and Altivec require
1919 // 16-byte alignment.
1920 // We calculate the correct alignment from the header before
1921 // reading the file, and then we misalign image on purpose so
1922 // that the actual sections end up aligned again.
1923 misalignment = machoGetMisalignment(f);
1924 image = stgMallocBytes(fileSize + misalignment, "loadObj(image)");
1925 image += misalignment;
1927 image = stgMallocBytes(fileSize, "loadObj(image)");
1932 n = fread ( image, 1, fileSize, f );
1934 barf("loadObj: error whilst reading `%s'", path);
1937 #endif /* USE_MMAP */
1939 oc = mkOc(path, image, fileSize, NULL
1941 #ifdef darwin_HOST_OS
1951 loadOc( ObjectCode* oc ) {
1954 IF_DEBUG(linker, debugBelch("loadOc\n"));
1956 # if defined(OBJFORMAT_MACHO) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1957 r = ocAllocateSymbolExtras_MachO ( oc );
1959 IF_DEBUG(linker, debugBelch("ocAllocateSymbolExtras_MachO failed\n"));
1962 # elif defined(OBJFORMAT_ELF) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1963 r = ocAllocateSymbolExtras_ELF ( oc );
1965 IF_DEBUG(linker, debugBelch("ocAllocateSymbolExtras_ELF failed\n"));
1970 /* verify the in-memory image */
1971 # if defined(OBJFORMAT_ELF)
1972 r = ocVerifyImage_ELF ( oc );
1973 # elif defined(OBJFORMAT_PEi386)
1974 r = ocVerifyImage_PEi386 ( oc );
1975 # elif defined(OBJFORMAT_MACHO)
1976 r = ocVerifyImage_MachO ( oc );
1978 barf("loadObj: no verify method");
1981 IF_DEBUG(linker, debugBelch("ocVerifyImage_* failed\n"));
1985 /* build the symbol list for this image */
1986 # if defined(OBJFORMAT_ELF)
1987 r = ocGetNames_ELF ( oc );
1988 # elif defined(OBJFORMAT_PEi386)
1989 r = ocGetNames_PEi386 ( oc );
1990 # elif defined(OBJFORMAT_MACHO)
1991 r = ocGetNames_MachO ( oc );
1993 barf("loadObj: no getNames method");
1996 IF_DEBUG(linker, debugBelch("ocGetNames_* failed\n"));
2000 /* loaded, but not resolved yet */
2001 oc->status = OBJECT_LOADED;
2002 IF_DEBUG(linker, debugBelch("loadObj done.\n"));
2007 /* -----------------------------------------------------------------------------
2008 * resolve all the currently unlinked objects in memory
2010 * Returns: 1 if ok, 0 on error.
2018 IF_DEBUG(linker, debugBelch("resolveObjs: start\n"));
2021 for (oc = objects; oc; oc = oc->next) {
2022 if (oc->status != OBJECT_RESOLVED) {
2023 # if defined(OBJFORMAT_ELF)
2024 r = ocResolve_ELF ( oc );
2025 # elif defined(OBJFORMAT_PEi386)
2026 r = ocResolve_PEi386 ( oc );
2027 # elif defined(OBJFORMAT_MACHO)
2028 r = ocResolve_MachO ( oc );
2030 barf("resolveObjs: not implemented on this platform");
2032 if (!r) { return r; }
2033 oc->status = OBJECT_RESOLVED;
2036 IF_DEBUG(linker, debugBelch("resolveObjs: done\n"));
2040 /* -----------------------------------------------------------------------------
2041 * delete an object from the pool
2044 unloadObj( char *path )
2046 ObjectCode *oc, *prev;
2047 HsBool unloadedAnyObj = HS_BOOL_FALSE;
2049 ASSERT(symhash != NULL);
2050 ASSERT(objects != NULL);
2055 for (oc = objects; oc; prev = oc, oc = oc->next) {
2056 if (!strcmp(oc->fileName,path)) {
2058 /* Remove all the mappings for the symbols within this
2063 for (i = 0; i < oc->n_symbols; i++) {
2064 if (oc->symbols[i] != NULL) {
2065 removeStrHashTable(symhash, oc->symbols[i], NULL);
2073 prev->next = oc->next;
2076 // We're going to leave this in place, in case there are
2077 // any pointers from the heap into it:
2078 // #ifdef mingw32_HOST_OS
2079 // VirtualFree(oc->image);
2081 // stgFree(oc->image);
2083 stgFree(oc->fileName);
2084 stgFree(oc->symbols);
2085 stgFree(oc->sections);
2088 /* This could be a member of an archive so continue
2089 * unloading other members. */
2090 unloadedAnyObj = HS_BOOL_TRUE;
2094 if (unloadedAnyObj) {
2098 errorBelch("unloadObj: can't find `%s' to unload", path);
2103 /* -----------------------------------------------------------------------------
2104 * Sanity checking. For each ObjectCode, maintain a list of address ranges
2105 * which may be prodded during relocation, and abort if we try and write
2106 * outside any of these.
2108 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
2111 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
2112 IF_DEBUG(linker, debugBelch("addProddableBlock %p %p %d\n", oc, start, size));
2116 pb->next = oc->proddables;
2117 oc->proddables = pb;
2120 static void checkProddableBlock ( ObjectCode* oc, void* addr )
2123 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
2124 char* s = (char*)(pb->start);
2125 char* e = s + pb->size - 1;
2126 char* a = (char*)addr;
2127 /* Assumes that the biggest fixup involves a 4-byte write. This
2128 probably needs to be changed to 8 (ie, +7) on 64-bit
2130 if (a >= s && (a+3) <= e) return;
2132 barf("checkProddableBlock: invalid fixup in runtime linker");
2135 /* -----------------------------------------------------------------------------
2136 * Section management.
2138 static void addSection ( ObjectCode* oc, SectionKind kind,
2139 void* start, void* end )
2141 Section* s = stgMallocBytes(sizeof(Section), "addSection");
2145 s->next = oc->sections;
2148 debugBelch("addSection: %p-%p (size %d), kind %d\n",
2149 start, ((char*)end)-1, end - start + 1, kind );
2154 /* --------------------------------------------------------------------------
2156 * This is about allocating a small chunk of memory for every symbol in the
2157 * object file. We make sure that the SymboLExtras are always "in range" of
2158 * limited-range PC-relative instructions on various platforms by allocating
2159 * them right next to the object code itself.
2162 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
2165 ocAllocateSymbolExtras
2167 Allocate additional space at the end of the object file image to make room
2168 for jump islands (powerpc, x86_64) and GOT entries (x86_64).
2170 PowerPC relative branch instructions have a 24 bit displacement field.
2171 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
2172 If a particular imported symbol is outside this range, we have to redirect
2173 the jump to a short piece of new code that just loads the 32bit absolute
2174 address and jumps there.
2175 On x86_64, PC-relative jumps and PC-relative accesses to the GOT are limited
2178 This function just allocates space for one SymbolExtra for every
2179 undefined symbol in the object file. The code for the jump islands is
2180 filled in by makeSymbolExtra below.
2183 static int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first )
2190 int misalignment = 0;
2191 #ifdef darwin_HOST_OS
2192 misalignment = oc->misalignment;
2198 // round up to the nearest 4
2199 aligned = (oc->fileSize + 3) & ~3;
2202 pagesize = getpagesize();
2203 n = ROUND_UP( oc->fileSize, pagesize );
2204 m = ROUND_UP( aligned + sizeof (SymbolExtra) * count, pagesize );
2206 /* we try to use spare space at the end of the last page of the
2207 * image for the jump islands, but if there isn't enough space
2208 * then we have to map some (anonymously, remembering MAP_32BIT).
2210 if( m > n ) // we need to allocate more pages
2212 oc->symbol_extras = mmapForLinker(sizeof(SymbolExtra) * count,
2217 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
2220 oc->image -= misalignment;
2221 oc->image = stgReallocBytes( oc->image,
2223 aligned + sizeof (SymbolExtra) * count,
2224 "ocAllocateSymbolExtras" );
2225 oc->image += misalignment;
2227 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
2228 #endif /* USE_MMAP */
2230 memset( oc->symbol_extras, 0, sizeof (SymbolExtra) * count );
2233 oc->symbol_extras = NULL;
2235 oc->first_symbol_extra = first;
2236 oc->n_symbol_extras = count;
2241 static SymbolExtra* makeSymbolExtra( ObjectCode* oc,
2242 unsigned long symbolNumber,
2243 unsigned long target )
2247 ASSERT( symbolNumber >= oc->first_symbol_extra
2248 && symbolNumber - oc->first_symbol_extra < oc->n_symbol_extras);
2250 extra = &oc->symbol_extras[symbolNumber - oc->first_symbol_extra];
2252 #ifdef powerpc_HOST_ARCH
2253 // lis r12, hi16(target)
2254 extra->jumpIsland.lis_r12 = 0x3d80;
2255 extra->jumpIsland.hi_addr = target >> 16;
2257 // ori r12, r12, lo16(target)
2258 extra->jumpIsland.ori_r12_r12 = 0x618c;
2259 extra->jumpIsland.lo_addr = target & 0xffff;
2262 extra->jumpIsland.mtctr_r12 = 0x7d8903a6;
2265 extra->jumpIsland.bctr = 0x4e800420;
2267 #ifdef x86_64_HOST_ARCH
2269 static uint8_t jmp[] = { 0xFF, 0x25, 0xF2, 0xFF, 0xFF, 0xFF };
2270 extra->addr = target;
2271 memcpy(extra->jumpIsland, jmp, 6);
2279 /* --------------------------------------------------------------------------
2280 * PowerPC specifics (instruction cache flushing)
2281 * ------------------------------------------------------------------------*/
2283 #ifdef powerpc_HOST_ARCH
2285 ocFlushInstructionCache
2287 Flush the data & instruction caches.
2288 Because the PPC has split data/instruction caches, we have to
2289 do that whenever we modify code at runtime.
2292 static void ocFlushInstructionCache( ObjectCode *oc )
2294 int n = (oc->fileSize + sizeof( SymbolExtra ) * oc->n_symbol_extras + 3) / 4;
2295 unsigned long *p = (unsigned long *) oc->image;
2299 __asm__ volatile ( "dcbf 0,%0\n\t"
2307 __asm__ volatile ( "sync\n\t"
2313 /* --------------------------------------------------------------------------
2314 * PEi386 specifics (Win32 targets)
2315 * ------------------------------------------------------------------------*/
2317 /* The information for this linker comes from
2318 Microsoft Portable Executable
2319 and Common Object File Format Specification
2320 revision 5.1 January 1998
2321 which SimonM says comes from the MS Developer Network CDs.
2323 It can be found there (on older CDs), but can also be found
2326 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
2328 (this is Rev 6.0 from February 1999).
2330 Things move, so if that fails, try searching for it via
2332 http://www.google.com/search?q=PE+COFF+specification
2334 The ultimate reference for the PE format is the Winnt.h
2335 header file that comes with the Platform SDKs; as always,
2336 implementations will drift wrt their documentation.
2338 A good background article on the PE format is Matt Pietrek's
2339 March 1994 article in Microsoft System Journal (MSJ)
2340 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
2341 Win32 Portable Executable File Format." The info in there
2342 has recently been updated in a two part article in
2343 MSDN magazine, issues Feb and March 2002,
2344 "Inside Windows: An In-Depth Look into the Win32 Portable
2345 Executable File Format"
2347 John Levine's book "Linkers and Loaders" contains useful
2352 #if defined(OBJFORMAT_PEi386)
2356 typedef unsigned char UChar;
2357 typedef unsigned short UInt16;
2358 typedef unsigned int UInt32;
2365 UInt16 NumberOfSections;
2366 UInt32 TimeDateStamp;
2367 UInt32 PointerToSymbolTable;
2368 UInt32 NumberOfSymbols;
2369 UInt16 SizeOfOptionalHeader;
2370 UInt16 Characteristics;
2374 #define sizeof_COFF_header 20
2381 UInt32 VirtualAddress;
2382 UInt32 SizeOfRawData;
2383 UInt32 PointerToRawData;
2384 UInt32 PointerToRelocations;
2385 UInt32 PointerToLinenumbers;
2386 UInt16 NumberOfRelocations;
2387 UInt16 NumberOfLineNumbers;
2388 UInt32 Characteristics;
2392 #define sizeof_COFF_section 40
2399 UInt16 SectionNumber;
2402 UChar NumberOfAuxSymbols;
2406 #define sizeof_COFF_symbol 18
2411 UInt32 VirtualAddress;
2412 UInt32 SymbolTableIndex;
2417 #define sizeof_COFF_reloc 10
2420 /* From PE spec doc, section 3.3.2 */
2421 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
2422 windows.h -- for the same purpose, but I want to know what I'm
2424 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
2425 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
2426 #define MYIMAGE_FILE_DLL 0x2000
2427 #define MYIMAGE_FILE_SYSTEM 0x1000
2428 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
2429 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
2430 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
2432 /* From PE spec doc, section 5.4.2 and 5.4.4 */
2433 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
2434 #define MYIMAGE_SYM_CLASS_STATIC 3
2435 #define MYIMAGE_SYM_UNDEFINED 0
2437 /* From PE spec doc, section 4.1 */
2438 #define MYIMAGE_SCN_CNT_CODE 0x00000020
2439 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
2440 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
2442 /* From PE spec doc, section 5.2.1 */
2443 #define MYIMAGE_REL_I386_DIR32 0x0006
2444 #define MYIMAGE_REL_I386_REL32 0x0014
2447 /* We use myindex to calculate array addresses, rather than
2448 simply doing the normal subscript thing. That's because
2449 some of the above structs have sizes which are not
2450 a whole number of words. GCC rounds their sizes up to a
2451 whole number of words, which means that the address calcs
2452 arising from using normal C indexing or pointer arithmetic
2453 are just plain wrong. Sigh.
2456 myindex ( int scale, void* base, int index )
2459 ((UChar*)base) + scale * index;
2464 printName ( UChar* name, UChar* strtab )
2466 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2467 UInt32 strtab_offset = * (UInt32*)(name+4);
2468 debugBelch("%s", strtab + strtab_offset );
2471 for (i = 0; i < 8; i++) {
2472 if (name[i] == 0) break;
2473 debugBelch("%c", name[i] );
2480 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
2482 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2483 UInt32 strtab_offset = * (UInt32*)(name+4);
2484 strncpy ( (char*)dst, (char*)strtab+strtab_offset, dstSize );
2490 if (name[i] == 0) break;
2500 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
2503 /* If the string is longer than 8 bytes, look in the
2504 string table for it -- this will be correctly zero terminated.
2506 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2507 UInt32 strtab_offset = * (UInt32*)(name+4);
2508 return ((UChar*)strtab) + strtab_offset;
2510 /* Otherwise, if shorter than 8 bytes, return the original,
2511 which by defn is correctly terminated.
2513 if (name[7]==0) return name;
2514 /* The annoying case: 8 bytes. Copy into a temporary
2515 (XXX which is never freed ...)
2517 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
2519 strncpy((char*)newstr,(char*)name,8);
2524 /* Getting the name of a section is mildly tricky, so we make a
2525 function for it. Sadly, in one case we have to copy the string
2526 (when it is exactly 8 bytes long there's no trailing '\0'), so for
2527 consistency we *always* copy the string; the caller must free it
2530 cstring_from_section_name (UChar* name, UChar* strtab)
2535 int strtab_offset = strtol((char*)name+1,NULL,10);
2536 int len = strlen(((char*)strtab) + strtab_offset);
2538 newstr = stgMallocBytes(len, "cstring_from_section_symbol_name");
2539 strcpy((char*)newstr, (char*)((UChar*)strtab) + strtab_offset);
2544 newstr = stgMallocBytes(9, "cstring_from_section_symbol_name");
2546 strncpy((char*)newstr,(char*)name,8);
2552 /* Just compares the short names (first 8 chars) */
2553 static COFF_section *
2554 findPEi386SectionCalled ( ObjectCode* oc, UChar* name )
2558 = (COFF_header*)(oc->image);
2559 COFF_section* sectab
2561 ((UChar*)(oc->image))
2562 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2564 for (i = 0; i < hdr->NumberOfSections; i++) {
2567 COFF_section* section_i
2569 myindex ( sizeof_COFF_section, sectab, i );
2570 n1 = (UChar*) &(section_i->Name);
2572 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
2573 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
2574 n1[6]==n2[6] && n1[7]==n2[7])
2583 zapTrailingAtSign ( UChar* sym )
2585 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
2587 if (sym[0] == 0) return;
2589 while (sym[i] != 0) i++;
2592 while (j > 0 && my_isdigit(sym[j])) j--;
2593 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
2598 lookupSymbolInDLLs ( UChar *lbl )
2603 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
2604 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
2606 if (lbl[0] == '_') {
2607 /* HACK: if the name has an initial underscore, try stripping
2608 it off & look that up first. I've yet to verify whether there's
2609 a Rule that governs whether an initial '_' *should always* be
2610 stripped off when mapping from import lib name to the DLL name.
2612 sym = GetProcAddress(o_dll->instance, (char*)(lbl+1));
2614 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
2618 sym = GetProcAddress(o_dll->instance, (char*)lbl);
2620 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
2629 ocVerifyImage_PEi386 ( ObjectCode* oc )
2634 COFF_section* sectab;
2635 COFF_symbol* symtab;
2637 /* debugBelch("\nLOADING %s\n", oc->fileName); */
2638 hdr = (COFF_header*)(oc->image);
2639 sectab = (COFF_section*) (
2640 ((UChar*)(oc->image))
2641 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2643 symtab = (COFF_symbol*) (
2644 ((UChar*)(oc->image))
2645 + hdr->PointerToSymbolTable
2647 strtab = ((UChar*)symtab)
2648 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2650 if (hdr->Machine != 0x14c) {
2651 errorBelch("%s: Not x86 PEi386", oc->fileName);
2654 if (hdr->SizeOfOptionalHeader != 0) {
2655 errorBelch("%s: PEi386 with nonempty optional header", oc->fileName);
2658 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
2659 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
2660 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
2661 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
2662 errorBelch("%s: Not a PEi386 object file", oc->fileName);
2665 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
2666 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
2667 errorBelch("%s: Invalid PEi386 word size or endiannness: %d",
2669 (int)(hdr->Characteristics));
2672 /* If the string table size is way crazy, this might indicate that
2673 there are more than 64k relocations, despite claims to the
2674 contrary. Hence this test. */
2675 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
2677 if ( (*(UInt32*)strtab) > 600000 ) {
2678 /* Note that 600k has no special significance other than being
2679 big enough to handle the almost-2MB-sized lumps that
2680 constitute HSwin32*.o. */
2681 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
2686 /* No further verification after this point; only debug printing. */
2688 IF_DEBUG(linker, i=1);
2689 if (i == 0) return 1;
2691 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
2692 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
2693 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
2696 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
2697 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
2698 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
2699 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
2700 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
2701 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
2702 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
2704 /* Print the section table. */
2706 for (i = 0; i < hdr->NumberOfSections; i++) {
2708 COFF_section* sectab_i
2710 myindex ( sizeof_COFF_section, sectab, i );
2717 printName ( sectab_i->Name, strtab );
2727 sectab_i->VirtualSize,
2728 sectab_i->VirtualAddress,
2729 sectab_i->SizeOfRawData,
2730 sectab_i->PointerToRawData,
2731 sectab_i->NumberOfRelocations,
2732 sectab_i->PointerToRelocations,
2733 sectab_i->PointerToRawData
2735 reltab = (COFF_reloc*) (
2736 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2739 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2740 /* If the relocation field (a short) has overflowed, the
2741 * real count can be found in the first reloc entry.
2743 * See Section 4.1 (last para) of the PE spec (rev6.0).
2745 COFF_reloc* rel = (COFF_reloc*)
2746 myindex ( sizeof_COFF_reloc, reltab, 0 );
2747 noRelocs = rel->VirtualAddress;
2750 noRelocs = sectab_i->NumberOfRelocations;
2754 for (; j < noRelocs; j++) {
2756 COFF_reloc* rel = (COFF_reloc*)
2757 myindex ( sizeof_COFF_reloc, reltab, j );
2759 " type 0x%-4x vaddr 0x%-8x name `",
2761 rel->VirtualAddress );
2762 sym = (COFF_symbol*)
2763 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
2764 /* Hmm..mysterious looking offset - what's it for? SOF */
2765 printName ( sym->Name, strtab -10 );
2772 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
2773 debugBelch("---START of string table---\n");
2774 for (i = 4; i < *(Int32*)strtab; i++) {
2776 debugBelch("\n"); else
2777 debugBelch("%c", strtab[i] );
2779 debugBelch("--- END of string table---\n");
2784 COFF_symbol* symtab_i;
2785 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2786 symtab_i = (COFF_symbol*)
2787 myindex ( sizeof_COFF_symbol, symtab, i );
2793 printName ( symtab_i->Name, strtab );
2802 (Int32)(symtab_i->SectionNumber),
2803 (UInt32)symtab_i->Type,
2804 (UInt32)symtab_i->StorageClass,
2805 (UInt32)symtab_i->NumberOfAuxSymbols
2807 i += symtab_i->NumberOfAuxSymbols;
2817 ocGetNames_PEi386 ( ObjectCode* oc )
2820 COFF_section* sectab;
2821 COFF_symbol* symtab;
2828 hdr = (COFF_header*)(oc->image);
2829 sectab = (COFF_section*) (
2830 ((UChar*)(oc->image))
2831 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2833 symtab = (COFF_symbol*) (
2834 ((UChar*)(oc->image))
2835 + hdr->PointerToSymbolTable
2837 strtab = ((UChar*)(oc->image))
2838 + hdr->PointerToSymbolTable
2839 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2841 /* Allocate space for any (local, anonymous) .bss sections. */
2843 for (i = 0; i < hdr->NumberOfSections; i++) {
2846 COFF_section* sectab_i
2848 myindex ( sizeof_COFF_section, sectab, i );
2850 char *secname = cstring_from_section_name(sectab_i->Name, strtab);
2852 if (0 != strcmp(secname, ".bss")) {
2859 /* sof 10/05: the PE spec text isn't too clear regarding what
2860 * the SizeOfRawData field is supposed to hold for object
2861 * file sections containing just uninitialized data -- for executables,
2862 * it is supposed to be zero; unclear what it's supposed to be
2863 * for object files. However, VirtualSize is guaranteed to be
2864 * zero for object files, which definitely suggests that SizeOfRawData
2865 * will be non-zero (where else would the size of this .bss section be
2866 * stored?) Looking at the COFF_section info for incoming object files,
2867 * this certainly appears to be the case.
2869 * => I suspect we've been incorrectly handling .bss sections in (relocatable)
2870 * object files up until now. This turned out to bite us with ghc-6.4.1's use
2871 * of gcc-3.4.x, which has started to emit initially-zeroed-out local 'static'
2872 * variable decls into to the .bss section. (The specific function in Q which
2873 * triggered this is libraries/base/cbits/dirUtils.c:__hscore_getFolderPath())
2875 if (sectab_i->VirtualSize == 0 && sectab_i->SizeOfRawData == 0) continue;
2876 /* This is a non-empty .bss section. Allocate zeroed space for
2877 it, and set its PointerToRawData field such that oc->image +
2878 PointerToRawData == addr_of_zeroed_space. */
2879 bss_sz = sectab_i->VirtualSize;
2880 if ( bss_sz < sectab_i->SizeOfRawData) { bss_sz = sectab_i->SizeOfRawData; }
2881 zspace = stgCallocBytes(1, bss_sz, "ocGetNames_PEi386(anonymous bss)");
2882 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
2883 addProddableBlock(oc, zspace, bss_sz);
2884 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
2887 /* Copy section information into the ObjectCode. */
2889 for (i = 0; i < hdr->NumberOfSections; i++) {
2895 = SECTIONKIND_OTHER;
2896 COFF_section* sectab_i
2898 myindex ( sizeof_COFF_section, sectab, i );
2900 char *secname = cstring_from_section_name(sectab_i->Name, strtab);
2902 IF_DEBUG(linker, debugBelch("section name = %s\n", secname ));
2905 /* I'm sure this is the Right Way to do it. However, the
2906 alternative of testing the sectab_i->Name field seems to
2907 work ok with Cygwin.
2909 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
2910 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
2911 kind = SECTIONKIND_CODE_OR_RODATA;
2914 if (0==strcmp(".text",(char*)secname) ||
2915 0==strcmp(".rdata",(char*)secname)||
2916 0==strcmp(".rodata",(char*)secname))
2917 kind = SECTIONKIND_CODE_OR_RODATA;
2918 if (0==strcmp(".data",(char*)secname) ||
2919 0==strcmp(".bss",(char*)secname))
2920 kind = SECTIONKIND_RWDATA;
2922 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
2923 sz = sectab_i->SizeOfRawData;
2924 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
2926 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
2927 end = start + sz - 1;
2929 if (kind == SECTIONKIND_OTHER
2930 /* Ignore sections called which contain stabs debugging
2932 && 0 != strcmp(".stab", (char*)secname)
2933 && 0 != strcmp(".stabstr", (char*)secname)
2934 /* ignore constructor section for now */
2935 && 0 != strcmp(".ctors", (char*)secname)
2936 /* ignore section generated from .ident */
2937 && 0!= strncmp(".debug", (char*)secname, 6)
2938 /* ignore unknown section that appeared in gcc 3.4.5(?) */
2939 && 0!= strcmp(".reloc", (char*)secname)
2940 && 0 != strcmp(".rdata$zzz", (char*)secname)
2942 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", secname, oc->fileName);
2947 if (kind != SECTIONKIND_OTHER && end >= start) {
2948 addSection(oc, kind, start, end);
2949 addProddableBlock(oc, start, end - start + 1);
2955 /* Copy exported symbols into the ObjectCode. */
2957 oc->n_symbols = hdr->NumberOfSymbols;
2958 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2959 "ocGetNames_PEi386(oc->symbols)");
2960 /* Call me paranoid; I don't care. */
2961 for (i = 0; i < oc->n_symbols; i++)
2962 oc->symbols[i] = NULL;
2966 COFF_symbol* symtab_i;
2967 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2968 symtab_i = (COFF_symbol*)
2969 myindex ( sizeof_COFF_symbol, symtab, i );
2973 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2974 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2975 /* This symbol is global and defined, viz, exported */
2976 /* for MYIMAGE_SYMCLASS_EXTERNAL
2977 && !MYIMAGE_SYM_UNDEFINED,
2978 the address of the symbol is:
2979 address of relevant section + offset in section
2981 COFF_section* sectabent
2982 = (COFF_section*) myindex ( sizeof_COFF_section,
2984 symtab_i->SectionNumber-1 );
2985 addr = ((UChar*)(oc->image))
2986 + (sectabent->PointerToRawData
2990 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2991 && symtab_i->Value > 0) {
2992 /* This symbol isn't in any section at all, ie, global bss.
2993 Allocate zeroed space for it. */
2994 addr = stgCallocBytes(1, symtab_i->Value,
2995 "ocGetNames_PEi386(non-anonymous bss)");
2996 addSection(oc, SECTIONKIND_RWDATA, addr,
2997 ((UChar*)addr) + symtab_i->Value - 1);
2998 addProddableBlock(oc, addr, symtab_i->Value);
2999 /* debugBelch("BSS section at 0x%x\n", addr); */
3002 if (addr != NULL ) {
3003 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
3004 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
3005 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
3006 ASSERT(i >= 0 && i < oc->n_symbols);
3007 /* cstring_from_COFF_symbol_name always succeeds. */
3008 oc->symbols[i] = (char*)sname;
3009 ghciInsertStrHashTable(oc->fileName, symhash, (char*)sname, addr);
3013 "IGNORING symbol %d\n"
3017 printName ( symtab_i->Name, strtab );
3026 (Int32)(symtab_i->SectionNumber),
3027 (UInt32)symtab_i->Type,
3028 (UInt32)symtab_i->StorageClass,
3029 (UInt32)symtab_i->NumberOfAuxSymbols
3034 i += symtab_i->NumberOfAuxSymbols;
3043 ocResolve_PEi386 ( ObjectCode* oc )
3046 COFF_section* sectab;
3047 COFF_symbol* symtab;
3057 /* ToDo: should be variable-sized? But is at least safe in the
3058 sense of buffer-overrun-proof. */
3060 /* debugBelch("resolving for %s\n", oc->fileName); */
3062 hdr = (COFF_header*)(oc->image);
3063 sectab = (COFF_section*) (
3064 ((UChar*)(oc->image))
3065 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
3067 symtab = (COFF_symbol*) (
3068 ((UChar*)(oc->image))
3069 + hdr->PointerToSymbolTable
3071 strtab = ((UChar*)(oc->image))
3072 + hdr->PointerToSymbolTable
3073 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
3075 for (i = 0; i < hdr->NumberOfSections; i++) {
3076 COFF_section* sectab_i
3078 myindex ( sizeof_COFF_section, sectab, i );
3081 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
3084 char *secname = cstring_from_section_name(sectab_i->Name, strtab);
3086 /* Ignore sections called which contain stabs debugging
3088 if (0 == strcmp(".stab", (char*)secname)
3089 || 0 == strcmp(".stabstr", (char*)secname)
3090 || 0 == strcmp(".ctors", (char*)secname)
3091 || 0 == strncmp(".debug", (char*)secname, 6)
3092 || 0 == strcmp(".rdata$zzz", (char*)secname)) {
3099 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
3100 /* If the relocation field (a short) has overflowed, the
3101 * real count can be found in the first reloc entry.
3103 * See Section 4.1 (last para) of the PE spec (rev6.0).
3105 * Nov2003 update: the GNU linker still doesn't correctly
3106 * handle the generation of relocatable object files with
3107 * overflown relocations. Hence the output to warn of potential
3110 COFF_reloc* rel = (COFF_reloc*)
3111 myindex ( sizeof_COFF_reloc, reltab, 0 );
3112 noRelocs = rel->VirtualAddress;
3114 /* 10/05: we now assume (and check for) a GNU ld that is capable
3115 * of handling object files with (>2^16) of relocs.
3118 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
3123 noRelocs = sectab_i->NumberOfRelocations;
3128 for (; j < noRelocs; j++) {
3130 COFF_reloc* reltab_j
3132 myindex ( sizeof_COFF_reloc, reltab, j );
3134 /* the location to patch */
3136 ((UChar*)(oc->image))
3137 + (sectab_i->PointerToRawData
3138 + reltab_j->VirtualAddress
3139 - sectab_i->VirtualAddress )
3141 /* the existing contents of pP */
3143 /* the symbol to connect to */
3144 sym = (COFF_symbol*)
3145 myindex ( sizeof_COFF_symbol,
3146 symtab, reltab_j->SymbolTableIndex );
3149 "reloc sec %2d num %3d: type 0x%-4x "
3150 "vaddr 0x%-8x name `",
3152 (UInt32)reltab_j->Type,
3153 reltab_j->VirtualAddress );
3154 printName ( sym->Name, strtab );
3155 debugBelch("'\n" ));
3157 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
3158 COFF_section* section_sym
3159 = findPEi386SectionCalled ( oc, sym->Name );
3161 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
3164 S = ((UInt32)(oc->image))
3165 + (section_sym->PointerToRawData
3168 copyName ( sym->Name, strtab, symbol, 1000-1 );
3169 S = (UInt32) lookupSymbol( (char*)symbol );
3170 if ((void*)S != NULL) goto foundit;
3171 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3175 checkProddableBlock(oc, pP);
3176 switch (reltab_j->Type) {
3177 case MYIMAGE_REL_I386_DIR32:
3180 case MYIMAGE_REL_I386_REL32:
3181 /* Tricky. We have to insert a displacement at
3182 pP which, when added to the PC for the _next_
3183 insn, gives the address of the target (S).
3184 Problem is to know the address of the next insn
3185 when we only know pP. We assume that this
3186 literal field is always the last in the insn,
3187 so that the address of the next insn is pP+4
3188 -- hence the constant 4.
3189 Also I don't know if A should be added, but so
3190 far it has always been zero.
3192 SOF 05/2005: 'A' (old contents of *pP) have been observed
3193 to contain values other than zero (the 'wx' object file
3194 that came with wxhaskell-0.9.4; dunno how it was compiled..).
3195 So, add displacement to old value instead of asserting
3196 A to be zero. Fixes wxhaskell-related crashes, and no other
3197 ill effects have been observed.
3199 Update: the reason why we're seeing these more elaborate
3200 relocations is due to a switch in how the NCG compiles SRTs
3201 and offsets to them from info tables. SRTs live in .(ro)data,
3202 while info tables live in .text, causing GAS to emit REL32/DISP32
3203 relocations with non-zero values. Adding the displacement is
3204 the right thing to do.
3206 *pP = S - ((UInt32)pP) - 4 + A;
3209 debugBelch("%s: unhandled PEi386 relocation type %d",
3210 oc->fileName, reltab_j->Type);
3217 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
3221 #endif /* defined(OBJFORMAT_PEi386) */
3224 /* --------------------------------------------------------------------------
3226 * ------------------------------------------------------------------------*/
3228 #if defined(OBJFORMAT_ELF)
3233 #if defined(sparc_HOST_ARCH)
3234 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
3235 #elif defined(i386_HOST_ARCH)
3236 # define ELF_TARGET_386 /* Used inside <elf.h> */
3237 #elif defined(x86_64_HOST_ARCH)
3238 # define ELF_TARGET_X64_64
3242 #if !defined(openbsd_HOST_OS)
3245 /* openbsd elf has things in different places, with diff names */
3246 # include <elf_abi.h>
3247 # include <machine/reloc.h>
3248 # define R_386_32 RELOC_32
3249 # define R_386_PC32 RELOC_PC32
3252 /* If elf.h doesn't define it */
3253 # ifndef R_X86_64_PC64
3254 # define R_X86_64_PC64 24
3258 * Define a set of types which can be used for both ELF32 and ELF64
3262 #define ELFCLASS ELFCLASS64
3263 #define Elf_Addr Elf64_Addr
3264 #define Elf_Word Elf64_Word
3265 #define Elf_Sword Elf64_Sword
3266 #define Elf_Ehdr Elf64_Ehdr
3267 #define Elf_Phdr Elf64_Phdr
3268 #define Elf_Shdr Elf64_Shdr
3269 #define Elf_Sym Elf64_Sym
3270 #define Elf_Rel Elf64_Rel
3271 #define Elf_Rela Elf64_Rela
3273 #define ELF_ST_TYPE ELF64_ST_TYPE
3276 #define ELF_ST_BIND ELF64_ST_BIND
3279 #define ELF_R_TYPE ELF64_R_TYPE
3282 #define ELF_R_SYM ELF64_R_SYM
3285 #define ELFCLASS ELFCLASS32
3286 #define Elf_Addr Elf32_Addr
3287 #define Elf_Word Elf32_Word
3288 #define Elf_Sword Elf32_Sword
3289 #define Elf_Ehdr Elf32_Ehdr
3290 #define Elf_Phdr Elf32_Phdr
3291 #define Elf_Shdr Elf32_Shdr
3292 #define Elf_Sym Elf32_Sym
3293 #define Elf_Rel Elf32_Rel
3294 #define Elf_Rela Elf32_Rela
3296 #define ELF_ST_TYPE ELF32_ST_TYPE
3299 #define ELF_ST_BIND ELF32_ST_BIND
3302 #define ELF_R_TYPE ELF32_R_TYPE
3305 #define ELF_R_SYM ELF32_R_SYM
3311 * Functions to allocate entries in dynamic sections. Currently we simply
3312 * preallocate a large number, and we don't check if a entry for the given
3313 * target already exists (a linear search is too slow). Ideally these
3314 * entries would be associated with symbols.
3317 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
3318 #define GOT_SIZE 0x20000
3319 #define FUNCTION_TABLE_SIZE 0x10000
3320 #define PLT_SIZE 0x08000
3323 static Elf_Addr got[GOT_SIZE];
3324 static unsigned int gotIndex;
3325 static Elf_Addr gp_val = (Elf_Addr)got;
3328 allocateGOTEntry(Elf_Addr target)
3332 if (gotIndex >= GOT_SIZE)
3333 barf("Global offset table overflow");
3335 entry = &got[gotIndex++];
3337 return (Elf_Addr)entry;
3341 #ifdef ELF_FUNCTION_DESC
3347 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
3348 static unsigned int functionTableIndex;
3351 allocateFunctionDesc(Elf_Addr target)
3353 FunctionDesc *entry;
3355 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
3356 barf("Function table overflow");
3358 entry = &functionTable[functionTableIndex++];
3360 entry->gp = (Elf_Addr)gp_val;
3361 return (Elf_Addr)entry;
3365 copyFunctionDesc(Elf_Addr target)
3367 FunctionDesc *olddesc = (FunctionDesc *)target;
3368 FunctionDesc *newdesc;
3370 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
3371 newdesc->gp = olddesc->gp;
3372 return (Elf_Addr)newdesc;
3379 unsigned char code[sizeof(plt_code)];
3383 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
3385 PLTEntry *plt = (PLTEntry *)oc->plt;
3388 if (oc->pltIndex >= PLT_SIZE)
3389 barf("Procedure table overflow");
3391 entry = &plt[oc->pltIndex++];
3392 memcpy(entry->code, plt_code, sizeof(entry->code));
3393 PLT_RELOC(entry->code, target);
3394 return (Elf_Addr)entry;
3400 return (PLT_SIZE * sizeof(PLTEntry));
3406 * Generic ELF functions
3410 findElfSection ( void* objImage, Elf_Word sh_type )
3412 char* ehdrC = (char*)objImage;
3413 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3414 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
3415 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3419 for (i = 0; i < ehdr->e_shnum; i++) {
3420 if (shdr[i].sh_type == sh_type
3421 /* Ignore the section header's string table. */
3422 && i != ehdr->e_shstrndx
3423 /* Ignore string tables named .stabstr, as they contain
3425 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3427 ptr = ehdrC + shdr[i].sh_offset;
3435 ocVerifyImage_ELF ( ObjectCode* oc )
3439 int i, j, nent, nstrtab, nsymtabs;
3443 char* ehdrC = (char*)(oc->image);
3444 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3446 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
3447 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
3448 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
3449 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
3450 errorBelch("%s: not an ELF object", oc->fileName);
3454 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
3455 errorBelch("%s: unsupported ELF format", oc->fileName);
3459 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
3460 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
3462 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
3463 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
3465 errorBelch("%s: unknown endiannness", oc->fileName);
3469 if (ehdr->e_type != ET_REL) {
3470 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
3473 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
3475 IF_DEBUG(linker,debugBelch( "Architecture is " ));
3476 switch (ehdr->e_machine) {
3477 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
3478 #ifdef EM_SPARC32PLUS
3479 case EM_SPARC32PLUS:
3481 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
3483 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
3485 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
3487 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
3488 #elif defined(EM_AMD64)
3489 case EM_AMD64: IF_DEBUG(linker,debugBelch( "amd64" )); break;
3491 default: IF_DEBUG(linker,debugBelch( "unknown" ));
3492 errorBelch("%s: unknown architecture (e_machine == %d)"
3493 , oc->fileName, ehdr->e_machine);
3497 IF_DEBUG(linker,debugBelch(
3498 "\nSection header table: start %ld, n_entries %d, ent_size %d\n",
3499 (long)ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
3501 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
3503 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3505 if (ehdr->e_shstrndx == SHN_UNDEF) {
3506 errorBelch("%s: no section header string table", oc->fileName);
3509 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
3511 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3514 for (i = 0; i < ehdr->e_shnum; i++) {
3515 IF_DEBUG(linker,debugBelch("%2d: ", i ));
3516 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
3517 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
3518 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
3519 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
3520 ehdrC + shdr[i].sh_offset,
3521 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
3523 if (shdr[i].sh_type == SHT_REL) {
3524 IF_DEBUG(linker,debugBelch("Rel " ));
3525 } else if (shdr[i].sh_type == SHT_RELA) {
3526 IF_DEBUG(linker,debugBelch("RelA " ));
3528 IF_DEBUG(linker,debugBelch(" "));
3531 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
3535 IF_DEBUG(linker,debugBelch( "\nString tables" ));
3538 for (i = 0; i < ehdr->e_shnum; i++) {
3539 if (shdr[i].sh_type == SHT_STRTAB
3540 /* Ignore the section header's string table. */
3541 && i != ehdr->e_shstrndx
3542 /* Ignore string tables named .stabstr, as they contain
3544 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3546 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
3547 strtab = ehdrC + shdr[i].sh_offset;
3552 errorBelch("%s: no string tables, or too many", oc->fileName);
3557 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
3558 for (i = 0; i < ehdr->e_shnum; i++) {
3559 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3560 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
3562 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3563 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3564 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%ld rem)\n",
3566 (long)shdr[i].sh_size % sizeof(Elf_Sym)
3568 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
3569 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
3572 for (j = 0; j < nent; j++) {
3573 IF_DEBUG(linker,debugBelch(" %2d ", j ));
3574 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
3575 (int)stab[j].st_shndx,
3576 (int)stab[j].st_size,
3577 (char*)stab[j].st_value ));
3579 IF_DEBUG(linker,debugBelch("type=" ));
3580 switch (ELF_ST_TYPE(stab[j].st_info)) {
3581 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
3582 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
3583 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
3584 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
3585 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
3586 default: IF_DEBUG(linker,debugBelch("? " )); break;
3588 IF_DEBUG(linker,debugBelch(" " ));
3590 IF_DEBUG(linker,debugBelch("bind=" ));
3591 switch (ELF_ST_BIND(stab[j].st_info)) {
3592 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
3593 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
3594 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
3595 default: IF_DEBUG(linker,debugBelch("? " )); break;
3597 IF_DEBUG(linker,debugBelch(" " ));
3599 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
3603 if (nsymtabs == 0) {
3604 errorBelch("%s: didn't find any symbol tables", oc->fileName);
3611 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
3615 if (hdr->sh_type == SHT_PROGBITS
3616 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
3617 /* .text-style section */
3618 return SECTIONKIND_CODE_OR_RODATA;
3621 if (hdr->sh_type == SHT_PROGBITS
3622 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3623 /* .data-style section */
3624 return SECTIONKIND_RWDATA;
3627 if (hdr->sh_type == SHT_PROGBITS
3628 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
3629 /* .rodata-style section */
3630 return SECTIONKIND_CODE_OR_RODATA;
3633 if (hdr->sh_type == SHT_NOBITS
3634 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3635 /* .bss-style section */
3637 return SECTIONKIND_RWDATA;
3640 return SECTIONKIND_OTHER;
3645 ocGetNames_ELF ( ObjectCode* oc )
3650 char* ehdrC = (char*)(oc->image);
3651 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3652 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
3653 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3655 ASSERT(symhash != NULL);
3658 errorBelch("%s: no strtab", oc->fileName);
3663 for (i = 0; i < ehdr->e_shnum; i++) {
3664 /* Figure out what kind of section it is. Logic derived from
3665 Figure 1.14 ("Special Sections") of the ELF document
3666 ("Portable Formats Specification, Version 1.1"). */
3668 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
3670 if (is_bss && shdr[i].sh_size > 0) {
3671 /* This is a non-empty .bss section. Allocate zeroed space for
3672 it, and set its .sh_offset field such that
3673 ehdrC + .sh_offset == addr_of_zeroed_space. */
3674 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
3675 "ocGetNames_ELF(BSS)");
3676 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
3678 debugBelch("BSS section at 0x%x, size %d\n",
3679 zspace, shdr[i].sh_size);
3683 /* fill in the section info */
3684 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
3685 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
3686 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
3687 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
3690 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3692 /* copy stuff into this module's object symbol table */
3693 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3694 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3696 oc->n_symbols = nent;
3697 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3698 "ocGetNames_ELF(oc->symbols)");
3700 for (j = 0; j < nent; j++) {
3702 char isLocal = FALSE; /* avoids uninit-var warning */
3704 char* nm = strtab + stab[j].st_name;
3705 int secno = stab[j].st_shndx;
3707 /* Figure out if we want to add it; if so, set ad to its
3708 address. Otherwise leave ad == NULL. */
3710 if (secno == SHN_COMMON) {
3712 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
3714 debugBelch("COMMON symbol, size %d name %s\n",
3715 stab[j].st_size, nm);
3717 /* Pointless to do addProddableBlock() for this area,
3718 since the linker should never poke around in it. */
3721 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
3722 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
3724 /* and not an undefined symbol */
3725 && stab[j].st_shndx != SHN_UNDEF
3726 /* and not in a "special section" */
3727 && stab[j].st_shndx < SHN_LORESERVE
3729 /* and it's a not a section or string table or anything silly */
3730 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
3731 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
3732 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
3735 /* Section 0 is the undefined section, hence > and not >=. */
3736 ASSERT(secno > 0 && secno < ehdr->e_shnum);
3738 if (shdr[secno].sh_type == SHT_NOBITS) {
3739 debugBelch(" BSS symbol, size %d off %d name %s\n",
3740 stab[j].st_size, stab[j].st_value, nm);
3743 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
3744 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
3747 #ifdef ELF_FUNCTION_DESC
3748 /* dlsym() and the initialisation table both give us function
3749 * descriptors, so to be consistent we store function descriptors
3750 * in the symbol table */
3751 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
3752 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
3754 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s\n",
3755 ad, oc->fileName, nm ));
3760 /* And the decision is ... */
3764 oc->symbols[j] = nm;
3767 /* Ignore entirely. */
3769 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
3773 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
3774 strtab + stab[j].st_name ));
3777 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
3778 (int)ELF_ST_BIND(stab[j].st_info),
3779 (int)ELF_ST_TYPE(stab[j].st_info),
3780 (int)stab[j].st_shndx,
3781 strtab + stab[j].st_name
3784 oc->symbols[j] = NULL;
3793 /* Do ELF relocations which lack an explicit addend. All x86-linux
3794 relocations appear to be of this form. */
3796 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
3797 Elf_Shdr* shdr, int shnum,
3798 Elf_Sym* stab, char* strtab )
3803 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
3804 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
3805 int target_shndx = shdr[shnum].sh_info;
3806 int symtab_shndx = shdr[shnum].sh_link;
3808 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3809 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
3810 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3811 target_shndx, symtab_shndx ));
3813 /* Skip sections that we're not interested in. */
3816 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
3817 if (kind == SECTIONKIND_OTHER) {
3818 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
3823 for (j = 0; j < nent; j++) {
3824 Elf_Addr offset = rtab[j].r_offset;
3825 Elf_Addr info = rtab[j].r_info;
3827 Elf_Addr P = ((Elf_Addr)targ) + offset;
3828 Elf_Word* pP = (Elf_Word*)P;
3833 StgStablePtr stablePtr;
3836 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
3837 j, (void*)offset, (void*)info ));
3839 IF_DEBUG(linker,debugBelch( " ZERO" ));
3842 Elf_Sym sym = stab[ELF_R_SYM(info)];
3843 /* First see if it is a local symbol. */
3844 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3845 /* Yes, so we can get the address directly from the ELF symbol
3847 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3849 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3850 + stab[ELF_R_SYM(info)].st_value);
3853 symbol = strtab + sym.st_name;
3854 stablePtr = (StgStablePtr)lookupHashTable(stablehash, (StgWord)symbol);
3855 if (NULL == stablePtr) {
3856 /* No, so look up the name in our global table. */
3857 S_tmp = lookupSymbol( symbol );
3858 S = (Elf_Addr)S_tmp;
3860 stableVal = deRefStablePtr( stablePtr );
3862 S = (Elf_Addr)S_tmp;
3866 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3869 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
3872 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
3873 (void*)P, (void*)S, (void*)A ));
3874 checkProddableBlock ( oc, pP );
3878 switch (ELF_R_TYPE(info)) {
3879 # ifdef i386_HOST_ARCH
3880 case R_386_32: *pP = value; break;
3881 case R_386_PC32: *pP = value - P; break;
3884 errorBelch("%s: unhandled ELF relocation(Rel) type %lu\n",
3885 oc->fileName, (lnat)ELF_R_TYPE(info));
3893 /* Do ELF relocations for which explicit addends are supplied.
3894 sparc-solaris relocations appear to be of this form. */
3896 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
3897 Elf_Shdr* shdr, int shnum,
3898 Elf_Sym* stab, char* strtab )
3901 char *symbol = NULL;
3903 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
3904 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
3905 int target_shndx = shdr[shnum].sh_info;
3906 int symtab_shndx = shdr[shnum].sh_link;
3908 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3909 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
3910 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3911 target_shndx, symtab_shndx ));
3913 for (j = 0; j < nent; j++) {
3914 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3915 /* This #ifdef only serves to avoid unused-var warnings. */
3916 Elf_Addr offset = rtab[j].r_offset;
3917 Elf_Addr P = targ + offset;
3919 Elf_Addr info = rtab[j].r_info;
3920 Elf_Addr A = rtab[j].r_addend;
3924 # if defined(sparc_HOST_ARCH)
3925 Elf_Word* pP = (Elf_Word*)P;
3927 # elif defined(powerpc_HOST_ARCH)
3931 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
3932 j, (void*)offset, (void*)info,
3935 IF_DEBUG(linker,debugBelch( " ZERO" ));
3938 Elf_Sym sym = stab[ELF_R_SYM(info)];
3939 /* First see if it is a local symbol. */
3940 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3941 /* Yes, so we can get the address directly from the ELF symbol
3943 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3945 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3946 + stab[ELF_R_SYM(info)].st_value);
3947 #ifdef ELF_FUNCTION_DESC
3948 /* Make a function descriptor for this function */
3949 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
3950 S = allocateFunctionDesc(S + A);
3955 /* No, so look up the name in our global table. */
3956 symbol = strtab + sym.st_name;
3957 S_tmp = lookupSymbol( symbol );
3958 S = (Elf_Addr)S_tmp;
3960 #ifdef ELF_FUNCTION_DESC
3961 /* If a function, already a function descriptor - we would
3962 have to copy it to add an offset. */
3963 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3964 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3968 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3971 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3974 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3975 (void*)P, (void*)S, (void*)A ));
3976 /* checkProddableBlock ( oc, (void*)P ); */
3980 switch (ELF_R_TYPE(info)) {
3981 # if defined(sparc_HOST_ARCH)
3982 case R_SPARC_WDISP30:
3983 w1 = *pP & 0xC0000000;
3984 w2 = (Elf_Word)((value - P) >> 2);
3985 ASSERT((w2 & 0xC0000000) == 0);
3990 w1 = *pP & 0xFFC00000;
3991 w2 = (Elf_Word)(value >> 10);
3992 ASSERT((w2 & 0xFFC00000) == 0);
3998 w2 = (Elf_Word)(value & 0x3FF);
3999 ASSERT((w2 & ~0x3FF) == 0);
4004 /* According to the Sun documentation:
4006 This relocation type resembles R_SPARC_32, except it refers to an
4007 unaligned word. That is, the word to be relocated must be treated
4008 as four separate bytes with arbitrary alignment, not as a word
4009 aligned according to the architecture requirements.
4012 w2 = (Elf_Word)value;
4014 // SPARC doesn't do misaligned writes of 32 bit words,
4015 // so we have to do this one byte-at-a-time.
4016 char *pPc = (char*)pP;
4017 pPc[0] = (char) ((Elf_Word)(w2 & 0xff000000) >> 24);
4018 pPc[1] = (char) ((Elf_Word)(w2 & 0x00ff0000) >> 16);
4019 pPc[2] = (char) ((Elf_Word)(w2 & 0x0000ff00) >> 8);
4020 pPc[3] = (char) ((Elf_Word)(w2 & 0x000000ff));
4024 w2 = (Elf_Word)value;
4027 # elif defined(powerpc_HOST_ARCH)
4028 case R_PPC_ADDR16_LO:
4029 *(Elf32_Half*) P = value;
4032 case R_PPC_ADDR16_HI:
4033 *(Elf32_Half*) P = value >> 16;
4036 case R_PPC_ADDR16_HA:
4037 *(Elf32_Half*) P = (value + 0x8000) >> 16;
4041 *(Elf32_Word *) P = value;
4045 *(Elf32_Word *) P = value - P;
4051 if( delta << 6 >> 6 != delta )
4053 value = (Elf_Addr) (&makeSymbolExtra( oc, ELF_R_SYM(info), value )
4057 if( value == 0 || delta << 6 >> 6 != delta )
4059 barf( "Unable to make SymbolExtra for #%d",
4065 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
4066 | (delta & 0x3fffffc);
4070 #if x86_64_HOST_ARCH
4072 *(Elf64_Xword *)P = value;
4077 #if defined(ALWAYS_PIC)
4078 barf("R_X86_64_PC32 relocation, but ALWAYS_PIC.");
4080 StgInt64 off = value - P;
4081 if (off >= 0x7fffffffL || off < -0x80000000L) {
4082 #if X86_64_ELF_NONPIC_HACK
4083 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
4085 off = pltAddress + A - P;
4087 barf("R_X86_64_PC32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
4088 symbol, off, oc->fileName );
4091 *(Elf64_Word *)P = (Elf64_Word)off;
4098 StgInt64 off = value - P;
4099 *(Elf64_Word *)P = (Elf64_Word)off;
4104 #if defined(ALWAYS_PIC)
4105 barf("R_X86_64_32 relocation, but ALWAYS_PIC.");
4107 if (value >= 0x7fffffffL) {
4108 #if X86_64_ELF_NONPIC_HACK
4109 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
4111 value = pltAddress + A;
4113 barf("R_X86_64_32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
4114 symbol, value, oc->fileName );
4117 *(Elf64_Word *)P = (Elf64_Word)value;
4122 #if defined(ALWAYS_PIC)
4123 barf("R_X86_64_32S relocation, but ALWAYS_PIC.");
4125 if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) {
4126 #if X86_64_ELF_NONPIC_HACK
4127 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
4129 value = pltAddress + A;
4131 barf("R_X86_64_32S relocation out of range: %s = %p\nRecompile %s with -fPIC.",
4132 symbol, value, oc->fileName );
4135 *(Elf64_Sword *)P = (Elf64_Sword)value;
4139 case R_X86_64_GOTPCREL:
4141 StgInt64 gotAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)->addr;
4142 StgInt64 off = gotAddress + A - P;
4143 *(Elf64_Word *)P = (Elf64_Word)off;
4147 case R_X86_64_PLT32:
4149 #if defined(ALWAYS_PIC)
4150 barf("R_X86_64_PLT32 relocation, but ALWAYS_PIC.");
4152 StgInt64 off = value - P;
4153 if (off >= 0x7fffffffL || off < -0x80000000L) {
4154 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
4156 off = pltAddress + A - P;
4158 *(Elf64_Word *)P = (Elf64_Word)off;
4165 errorBelch("%s: unhandled ELF relocation(RelA) type %lu\n",
4166 oc->fileName, (lnat)ELF_R_TYPE(info));
4175 ocResolve_ELF ( ObjectCode* oc )
4179 Elf_Sym* stab = NULL;
4180 char* ehdrC = (char*)(oc->image);
4181 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
4182 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
4184 /* first find "the" symbol table */
4185 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
4187 /* also go find the string table */
4188 strtab = findElfSection ( ehdrC, SHT_STRTAB );
4190 if (stab == NULL || strtab == NULL) {
4191 errorBelch("%s: can't find string or symbol table", oc->fileName);
4195 /* Process the relocation sections. */
4196 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
4197 if (shdr[shnum].sh_type == SHT_REL) {
4198 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
4199 shnum, stab, strtab );
4203 if (shdr[shnum].sh_type == SHT_RELA) {
4204 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
4205 shnum, stab, strtab );
4210 #if defined(powerpc_HOST_ARCH)
4211 ocFlushInstructionCache( oc );
4218 * PowerPC & X86_64 ELF specifics
4221 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
4223 static int ocAllocateSymbolExtras_ELF( ObjectCode *oc )
4229 ehdr = (Elf_Ehdr *) oc->image;
4230 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
4232 for( i = 0; i < ehdr->e_shnum; i++ )
4233 if( shdr[i].sh_type == SHT_SYMTAB )
4236 if( i == ehdr->e_shnum )
4238 errorBelch( "This ELF file contains no symtab" );
4242 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
4244 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
4245 (int) shdr[i].sh_entsize, (int) sizeof( Elf_Sym ) );
4250 return ocAllocateSymbolExtras( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
4253 #endif /* powerpc */
4257 /* --------------------------------------------------------------------------
4259 * ------------------------------------------------------------------------*/
4261 #if defined(OBJFORMAT_MACHO)
4264 Support for MachO linking on Darwin/MacOS X
4265 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
4267 I hereby formally apologize for the hackish nature of this code.
4268 Things that need to be done:
4269 *) implement ocVerifyImage_MachO
4270 *) add still more sanity checks.
4273 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
4274 #define mach_header mach_header_64
4275 #define segment_command segment_command_64
4276 #define section section_64
4277 #define nlist nlist_64
4280 #ifdef powerpc_HOST_ARCH
4281 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
4283 struct mach_header *header = (struct mach_header *) oc->image;
4284 struct load_command *lc = (struct load_command *) (header + 1);
4287 for( i = 0; i < header->ncmds; i++ )
4289 if( lc->cmd == LC_SYMTAB )
4291 // Find out the first and last undefined external
4292 // symbol, so we don't have to allocate too many
4294 struct symtab_command *symLC = (struct symtab_command *) lc;
4295 unsigned min = symLC->nsyms, max = 0;
4296 struct nlist *nlist =
4297 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
4299 for(i=0;i<symLC->nsyms;i++)
4301 if(nlist[i].n_type & N_STAB)
4303 else if(nlist[i].n_type & N_EXT)
4305 if((nlist[i].n_type & N_TYPE) == N_UNDF
4306 && (nlist[i].n_value == 0))
4316 return ocAllocateSymbolExtras(oc, max - min + 1, min);
4321 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
4323 return ocAllocateSymbolExtras(oc,0,0);
4326 #ifdef x86_64_HOST_ARCH
4327 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
4329 struct mach_header *header = (struct mach_header *) oc->image;
4330 struct load_command *lc = (struct load_command *) (header + 1);
4333 for( i = 0; i < header->ncmds; i++ )
4335 if( lc->cmd == LC_SYMTAB )
4337 // Just allocate one entry for every symbol
4338 struct symtab_command *symLC = (struct symtab_command *) lc;
4340 return ocAllocateSymbolExtras(oc, symLC->nsyms, 0);
4343 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
4345 return ocAllocateSymbolExtras(oc,0,0);
4349 static int ocVerifyImage_MachO(ObjectCode* oc)
4351 char *image = (char*) oc->image;
4352 struct mach_header *header = (struct mach_header*) image;
4354 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
4355 if(header->magic != MH_MAGIC_64) {
4356 errorBelch("%s: Bad magic. Expected: %08x, got: %08x.\n",
4357 oc->fileName, MH_MAGIC_64, header->magic);
4361 if(header->magic != MH_MAGIC) {
4362 errorBelch("%s: Bad magic. Expected: %08x, got: %08x.\n",
4363 oc->fileName, MH_MAGIC, header->magic);
4367 // FIXME: do some more verifying here
4371 static int resolveImports(
4374 struct symtab_command *symLC,
4375 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
4376 unsigned long *indirectSyms,
4377 struct nlist *nlist)
4380 size_t itemSize = 4;
4382 IF_DEBUG(linker, debugBelch("resolveImports: start\n"));
4385 int isJumpTable = 0;
4386 if(!strcmp(sect->sectname,"__jump_table"))
4390 ASSERT(sect->reserved2 == itemSize);
4394 for(i=0; i*itemSize < sect->size;i++)
4396 // according to otool, reserved1 contains the first index into the indirect symbol table
4397 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
4398 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4401 IF_DEBUG(linker, debugBelch("resolveImports: resolving %s\n", nm));
4402 if ((symbol->n_type & N_TYPE) == N_UNDF
4403 && (symbol->n_type & N_EXT) && (symbol->n_value != 0)) {
4404 addr = (void*) (symbol->n_value);
4405 IF_DEBUG(linker, debugBelch("resolveImports: undefined external %s has value %p\n", nm, addr));
4407 addr = lookupSymbol(nm);
4408 IF_DEBUG(linker, debugBelch("resolveImports: looking up %s, %p\n", nm, addr));
4412 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
4420 checkProddableBlock(oc,image + sect->offset + i*itemSize);
4421 *(image + sect->offset + i*itemSize) = 0xe9; // jmp
4422 *(unsigned*)(image + sect->offset + i*itemSize + 1)
4423 = (char*)addr - (image + sect->offset + i*itemSize + 5);
4428 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
4429 ((void**)(image + sect->offset))[i] = addr;
4433 IF_DEBUG(linker, debugBelch("resolveImports: done\n"));
4437 static unsigned long relocateAddress(
4440 struct section* sections,
4441 unsigned long address)
4444 IF_DEBUG(linker, debugBelch("relocateAddress: start\n"));
4445 for (i = 0; i < nSections; i++)
4447 IF_DEBUG(linker, debugBelch(" relocating address in section %d\n", i));
4448 if (sections[i].addr <= address
4449 && address < sections[i].addr + sections[i].size)
4451 return (unsigned long)oc->image
4452 + sections[i].offset + address - sections[i].addr;
4455 barf("Invalid Mach-O file:"
4456 "Address out of bounds while relocating object file");
4460 static int relocateSection(
4463 struct symtab_command *symLC, struct nlist *nlist,
4464 int nSections, struct section* sections, struct section *sect)
4466 struct relocation_info *relocs;
4469 IF_DEBUG(linker, debugBelch("relocateSection: start\n"));
4471 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
4473 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
4475 else if(!strcmp(sect->sectname,"__la_sym_ptr2"))
4477 else if(!strcmp(sect->sectname,"__la_sym_ptr3"))
4481 IF_DEBUG(linker, debugBelch("relocateSection: number of relocations: %d\n", n));
4483 relocs = (struct relocation_info*) (image + sect->reloff);
4487 #ifdef x86_64_HOST_ARCH
4488 struct relocation_info *reloc = &relocs[i];
4490 char *thingPtr = image + sect->offset + reloc->r_address;
4492 /* We shouldn't need to initialise this, but gcc on OS X 64 bit
4493 complains that it may be used uninitialized if we don't */
4496 int type = reloc->r_type;
4498 checkProddableBlock(oc,thingPtr);
4499 switch(reloc->r_length)
4502 thing = *(uint8_t*)thingPtr;
4503 baseValue = (uint64_t)thingPtr + 1;
4506 thing = *(uint16_t*)thingPtr;
4507 baseValue = (uint64_t)thingPtr + 2;
4510 thing = *(uint32_t*)thingPtr;
4511 baseValue = (uint64_t)thingPtr + 4;
4514 thing = *(uint64_t*)thingPtr;
4515 baseValue = (uint64_t)thingPtr + 8;
4518 barf("Unknown size.");
4522 debugBelch("relocateSection: length = %d, thing = %d, baseValue = %p\n",
4523 reloc->r_length, thing, baseValue));
4525 if (type == X86_64_RELOC_GOT
4526 || type == X86_64_RELOC_GOT_LOAD)
4528 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4529 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4531 IF_DEBUG(linker, debugBelch("relocateSection: making jump island for %s, extern = %d, X86_64_RELOC_GOT\n", nm, reloc->r_extern));
4532 ASSERT(reloc->r_extern);
4533 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, (unsigned long)lookupSymbol(nm))->addr;
4535 type = X86_64_RELOC_SIGNED;
4537 else if(reloc->r_extern)
4539 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4540 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4542 IF_DEBUG(linker, debugBelch("relocateSection: looking up external symbol %s\n", nm));
4543 IF_DEBUG(linker, debugBelch(" : type = %d\n", symbol->n_type));
4544 IF_DEBUG(linker, debugBelch(" : sect = %d\n", symbol->n_sect));
4545 IF_DEBUG(linker, debugBelch(" : desc = %d\n", symbol->n_desc));
4546 IF_DEBUG(linker, debugBelch(" : value = %d\n", symbol->n_value));
4547 if ((symbol->n_type & N_TYPE) == N_SECT) {
4548 value = relocateAddress(oc, nSections, sections,
4550 IF_DEBUG(linker, debugBelch("relocateSection, defined external symbol %s, relocated address %p\n", nm, value));
4553 value = (uint64_t) lookupSymbol(nm);
4554 IF_DEBUG(linker, debugBelch("relocateSection: external symbol %s, address %p\n", nm, value));
4559 value = sections[reloc->r_symbolnum-1].offset
4560 - sections[reloc->r_symbolnum-1].addr
4564 IF_DEBUG(linker, debugBelch("relocateSection: value = %p\n", value));
4566 if (type == X86_64_RELOC_BRANCH)
4568 if((int32_t)(value - baseValue) != (int64_t)(value - baseValue))
4570 ASSERT(reloc->r_extern);
4571 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)
4574 ASSERT((int32_t)(value - baseValue) == (int64_t)(value - baseValue));
4575 type = X86_64_RELOC_SIGNED;
4580 case X86_64_RELOC_UNSIGNED:
4581 ASSERT(!reloc->r_pcrel);
4584 case X86_64_RELOC_SIGNED:
4585 case X86_64_RELOC_SIGNED_1:
4586 case X86_64_RELOC_SIGNED_2:
4587 case X86_64_RELOC_SIGNED_4:
4588 ASSERT(reloc->r_pcrel);
4589 thing += value - baseValue;
4591 case X86_64_RELOC_SUBTRACTOR:
4592 ASSERT(!reloc->r_pcrel);
4596 barf("unkown relocation");
4599 switch(reloc->r_length)
4602 *(uint8_t*)thingPtr = thing;
4605 *(uint16_t*)thingPtr = thing;
4608 *(uint32_t*)thingPtr = thing;
4611 *(uint64_t*)thingPtr = thing;
4615 if(relocs[i].r_address & R_SCATTERED)
4617 struct scattered_relocation_info *scat =
4618 (struct scattered_relocation_info*) &relocs[i];
4622 if(scat->r_length == 2)
4624 unsigned long word = 0;
4625 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
4626 checkProddableBlock(oc,wordPtr);
4628 // Note on relocation types:
4629 // i386 uses the GENERIC_RELOC_* types,
4630 // while ppc uses special PPC_RELOC_* types.
4631 // *_RELOC_VANILLA and *_RELOC_PAIR have the same value
4632 // in both cases, all others are different.
4633 // Therefore, we use GENERIC_RELOC_VANILLA
4634 // and GENERIC_RELOC_PAIR instead of the PPC variants,
4635 // and use #ifdefs for the other types.
4637 // Step 1: Figure out what the relocated value should be
4638 if(scat->r_type == GENERIC_RELOC_VANILLA)
4640 word = *wordPtr + (unsigned long) relocateAddress(
4647 #ifdef powerpc_HOST_ARCH
4648 else if(scat->r_type == PPC_RELOC_SECTDIFF
4649 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
4650 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
4651 || scat->r_type == PPC_RELOC_HA16_SECTDIFF
4652 || scat->r_type == PPC_RELOC_LOCAL_SECTDIFF)
4654 else if(scat->r_type == GENERIC_RELOC_SECTDIFF
4655 || scat->r_type == GENERIC_RELOC_LOCAL_SECTDIFF)
4658 struct scattered_relocation_info *pair =
4659 (struct scattered_relocation_info*) &relocs[i+1];
4661 if(!pair->r_scattered || pair->r_type != GENERIC_RELOC_PAIR)
4662 barf("Invalid Mach-O file: "
4663 "RELOC_*_SECTDIFF not followed by RELOC_PAIR");
4665 word = (unsigned long)
4666 (relocateAddress(oc, nSections, sections, scat->r_value)
4667 - relocateAddress(oc, nSections, sections, pair->r_value));
4670 #ifdef powerpc_HOST_ARCH
4671 else if(scat->r_type == PPC_RELOC_HI16
4672 || scat->r_type == PPC_RELOC_LO16
4673 || scat->r_type == PPC_RELOC_HA16
4674 || scat->r_type == PPC_RELOC_LO14)
4675 { // these are generated by label+offset things
4676 struct relocation_info *pair = &relocs[i+1];
4677 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
4678 barf("Invalid Mach-O file: "
4679 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
4681 if(scat->r_type == PPC_RELOC_LO16)
4683 word = ((unsigned short*) wordPtr)[1];
4684 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4686 else if(scat->r_type == PPC_RELOC_LO14)
4688 barf("Unsupported Relocation: PPC_RELOC_LO14");
4689 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
4690 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4692 else if(scat->r_type == PPC_RELOC_HI16)
4694 word = ((unsigned short*) wordPtr)[1] << 16;
4695 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4697 else if(scat->r_type == PPC_RELOC_HA16)
4699 word = ((unsigned short*) wordPtr)[1] << 16;
4700 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4704 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
4712 barf ("Don't know how to handle this Mach-O "
4713 "scattered relocation entry: "
4714 "object file %s; entry type %ld; "
4716 OC_INFORMATIVE_FILENAME(oc),
4722 #ifdef powerpc_HOST_ARCH
4723 if(scat->r_type == GENERIC_RELOC_VANILLA
4724 || scat->r_type == PPC_RELOC_SECTDIFF)
4726 if(scat->r_type == GENERIC_RELOC_VANILLA
4727 || scat->r_type == GENERIC_RELOC_SECTDIFF
4728 || scat->r_type == GENERIC_RELOC_LOCAL_SECTDIFF)
4733 #ifdef powerpc_HOST_ARCH
4734 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
4736 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4738 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
4740 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4742 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
4744 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4745 + ((word & (1<<15)) ? 1 : 0);
4751 barf("Can't handle Mach-O scattered relocation entry "
4752 "with this r_length tag: "
4753 "object file %s; entry type %ld; "
4754 "r_length tag %ld; address %#lx\n",
4755 OC_INFORMATIVE_FILENAME(oc),
4762 else /* scat->r_pcrel */
4764 barf("Don't know how to handle *PC-relative* Mach-O "
4765 "scattered relocation entry: "
4766 "object file %s; entry type %ld; address %#lx\n",
4767 OC_INFORMATIVE_FILENAME(oc),
4774 else /* !(relocs[i].r_address & R_SCATTERED) */
4776 struct relocation_info *reloc = &relocs[i];
4777 if(reloc->r_pcrel && !reloc->r_extern)
4780 if(reloc->r_length == 2)
4782 unsigned long word = 0;
4783 #ifdef powerpc_HOST_ARCH
4784 unsigned long jumpIsland = 0;
4785 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
4786 // to avoid warning and to catch
4790 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
4791 checkProddableBlock(oc,wordPtr);
4793 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4797 #ifdef powerpc_HOST_ARCH
4798 else if(reloc->r_type == PPC_RELOC_LO16)
4800 word = ((unsigned short*) wordPtr)[1];
4801 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4803 else if(reloc->r_type == PPC_RELOC_HI16)
4805 word = ((unsigned short*) wordPtr)[1] << 16;
4806 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4808 else if(reloc->r_type == PPC_RELOC_HA16)
4810 word = ((unsigned short*) wordPtr)[1] << 16;
4811 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4813 else if(reloc->r_type == PPC_RELOC_BR24)
4816 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
4821 barf("Can't handle this Mach-O relocation entry "
4823 "object file %s; entry type %ld; address %#lx\n",
4824 OC_INFORMATIVE_FILENAME(oc),
4830 if(!reloc->r_extern)
4833 sections[reloc->r_symbolnum-1].offset
4834 - sections[reloc->r_symbolnum-1].addr
4841 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4842 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4843 void *symbolAddress = lookupSymbol(nm);
4846 errorBelch("\nunknown symbol `%s'", nm);
4852 #ifdef powerpc_HOST_ARCH
4853 // In the .o file, this should be a relative jump to NULL
4854 // and we'll change it to a relative jump to the symbol
4855 ASSERT(word + reloc->r_address == 0);
4856 jumpIsland = (unsigned long)
4857 &makeSymbolExtra(oc,
4859 (unsigned long) symbolAddress)
4863 offsetToJumpIsland = word + jumpIsland
4864 - (((long)image) + sect->offset - sect->addr);
4867 word += (unsigned long) symbolAddress
4868 - (((long)image) + sect->offset - sect->addr);
4872 word += (unsigned long) symbolAddress;
4876 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4881 #ifdef powerpc_HOST_ARCH
4882 else if(reloc->r_type == PPC_RELOC_LO16)
4884 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4887 else if(reloc->r_type == PPC_RELOC_HI16)
4889 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4892 else if(reloc->r_type == PPC_RELOC_HA16)
4894 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4895 + ((word & (1<<15)) ? 1 : 0);
4898 else if(reloc->r_type == PPC_RELOC_BR24)
4900 if((word & 0x03) != 0)
4901 barf("%s: unconditional relative branch with a displacement "
4902 "which isn't a multiple of 4 bytes: %#lx",
4903 OC_INFORMATIVE_FILENAME(oc),
4906 if((word & 0xFE000000) != 0xFE000000 &&
4907 (word & 0xFE000000) != 0x00000000)
4909 // The branch offset is too large.
4910 // Therefore, we try to use a jump island.
4913 barf("%s: unconditional relative branch out of range: "
4914 "no jump island available: %#lx",
4915 OC_INFORMATIVE_FILENAME(oc),
4919 word = offsetToJumpIsland;
4920 if((word & 0xFE000000) != 0xFE000000 &&
4921 (word & 0xFE000000) != 0x00000000)
4922 barf("%s: unconditional relative branch out of range: "
4923 "jump island out of range: %#lx",
4924 OC_INFORMATIVE_FILENAME(oc),
4927 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
4934 barf("Can't handle Mach-O relocation entry (not scattered) "
4935 "with this r_length tag: "
4936 "object file %s; entry type %ld; "
4937 "r_length tag %ld; address %#lx\n",
4938 OC_INFORMATIVE_FILENAME(oc),
4947 IF_DEBUG(linker, debugBelch("relocateSection: done\n"));
4951 static int ocGetNames_MachO(ObjectCode* oc)
4953 char *image = (char*) oc->image;
4954 struct mach_header *header = (struct mach_header*) image;
4955 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4956 unsigned i,curSymbol = 0;
4957 struct segment_command *segLC = NULL;
4958 struct section *sections;
4959 struct symtab_command *symLC = NULL;
4960 struct nlist *nlist;
4961 unsigned long commonSize = 0;
4962 char *commonStorage = NULL;
4963 unsigned long commonCounter;
4965 IF_DEBUG(linker,debugBelch("ocGetNames_MachO: start\n"));
4967 for(i=0;i<header->ncmds;i++)
4969 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4970 segLC = (struct segment_command*) lc;
4971 else if(lc->cmd == LC_SYMTAB)
4972 symLC = (struct symtab_command*) lc;
4973 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4976 sections = (struct section*) (segLC+1);
4977 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4981 barf("ocGetNames_MachO: no segment load command");
4983 for(i=0;i<segLC->nsects;i++)
4985 IF_DEBUG(linker, debugBelch("ocGetNames_MachO: segment %d\n", i));
4986 if (sections[i].size == 0)
4989 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
4991 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
4992 "ocGetNames_MachO(common symbols)");
4993 sections[i].offset = zeroFillArea - image;
4996 if(!strcmp(sections[i].sectname,"__text"))
4997 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
4998 (void*) (image + sections[i].offset),
4999 (void*) (image + sections[i].offset + sections[i].size));
5000 else if(!strcmp(sections[i].sectname,"__const"))
5001 addSection(oc, SECTIONKIND_RWDATA,
5002 (void*) (image + sections[i].offset),
5003 (void*) (image + sections[i].offset + sections[i].size));
5004 else if(!strcmp(sections[i].sectname,"__data"))
5005 addSection(oc, SECTIONKIND_RWDATA,
5006 (void*) (image + sections[i].offset),
5007 (void*) (image + sections[i].offset + sections[i].size));
5008 else if(!strcmp(sections[i].sectname,"__bss")
5009 || !strcmp(sections[i].sectname,"__common"))
5010 addSection(oc, SECTIONKIND_RWDATA,
5011 (void*) (image + sections[i].offset),
5012 (void*) (image + sections[i].offset + sections[i].size));
5014 addProddableBlock(oc, (void*) (image + sections[i].offset),
5018 // count external symbols defined here
5022 for(i=0;i<symLC->nsyms;i++)
5024 if(nlist[i].n_type & N_STAB)
5026 else if(nlist[i].n_type & N_EXT)
5028 if((nlist[i].n_type & N_TYPE) == N_UNDF
5029 && (nlist[i].n_value != 0))
5031 commonSize += nlist[i].n_value;
5034 else if((nlist[i].n_type & N_TYPE) == N_SECT)
5039 IF_DEBUG(linker, debugBelch("ocGetNames_MachO: %d external symbols\n", oc->n_symbols));
5040 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
5041 "ocGetNames_MachO(oc->symbols)");
5045 for(i=0;i<symLC->nsyms;i++)
5047 if(nlist[i].n_type & N_STAB)
5049 else if((nlist[i].n_type & N_TYPE) == N_SECT)
5051 if(nlist[i].n_type & N_EXT)
5053 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
5054 if ((nlist[i].n_desc & N_WEAK_DEF) && lookupSymbol(nm)) {
5055 // weak definition, and we already have a definition
5056 IF_DEBUG(linker, debugBelch(" weak: %s\n", nm));
5060 IF_DEBUG(linker, debugBelch("ocGetNames_MachO: inserting %s\n", nm));
5061 ghciInsertStrHashTable(oc->fileName, symhash, nm,
5063 + sections[nlist[i].n_sect-1].offset
5064 - sections[nlist[i].n_sect-1].addr
5065 + nlist[i].n_value);
5066 oc->symbols[curSymbol++] = nm;
5073 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
5074 commonCounter = (unsigned long)commonStorage;
5077 for(i=0;i<symLC->nsyms;i++)
5079 if((nlist[i].n_type & N_TYPE) == N_UNDF
5080 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
5082 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
5083 unsigned long sz = nlist[i].n_value;
5085 nlist[i].n_value = commonCounter;
5087 IF_DEBUG(linker, debugBelch("ocGetNames_MachO: inserting common symbol: %s\n", nm));
5088 ghciInsertStrHashTable(oc->fileName, symhash, nm,
5089 (void*)commonCounter);
5090 oc->symbols[curSymbol++] = nm;
5092 commonCounter += sz;
5099 static int ocResolve_MachO(ObjectCode* oc)
5101 char *image = (char*) oc->image;
5102 struct mach_header *header = (struct mach_header*) image;
5103 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
5105 struct segment_command *segLC = NULL;
5106 struct section *sections;
5107 struct symtab_command *symLC = NULL;
5108 struct dysymtab_command *dsymLC = NULL;
5109 struct nlist *nlist;
5111 IF_DEBUG(linker, debugBelch("ocResolve_MachO: start\n"));
5112 for (i = 0; i < header->ncmds; i++)
5114 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
5115 segLC = (struct segment_command*) lc;
5116 else if(lc->cmd == LC_SYMTAB)
5117 symLC = (struct symtab_command*) lc;
5118 else if(lc->cmd == LC_DYSYMTAB)
5119 dsymLC = (struct dysymtab_command*) lc;
5120 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
5123 sections = (struct section*) (segLC+1);
5124 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
5129 unsigned long *indirectSyms
5130 = (unsigned long*) (image + dsymLC->indirectsymoff);
5132 IF_DEBUG(linker, debugBelch("ocResolve_MachO: resolving dsymLC\n"));
5133 for (i = 0; i < segLC->nsects; i++)
5135 if( !strcmp(sections[i].sectname,"__la_symbol_ptr")
5136 || !strcmp(sections[i].sectname,"__la_sym_ptr2")
5137 || !strcmp(sections[i].sectname,"__la_sym_ptr3"))
5139 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
5142 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr")
5143 || !strcmp(sections[i].sectname,"__pointers"))
5145 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
5148 else if(!strcmp(sections[i].sectname,"__jump_table"))
5150 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
5155 IF_DEBUG(linker, debugBelch("ocResolve_MachO: unknown section\n"));
5160 for(i=0;i<segLC->nsects;i++)
5162 IF_DEBUG(linker, debugBelch("ocResolve_MachO: relocating section %d\n", i));
5164 if (!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
5168 #if defined (powerpc_HOST_ARCH)
5169 ocFlushInstructionCache( oc );
5175 #ifdef powerpc_HOST_ARCH
5177 * The Mach-O object format uses leading underscores. But not everywhere.
5178 * There is a small number of runtime support functions defined in
5179 * libcc_dynamic.a whose name does not have a leading underscore.
5180 * As a consequence, we can't get their address from C code.
5181 * We have to use inline assembler just to take the address of a function.
5185 extern void* symbolsWithoutUnderscore[];
5187 static void machoInitSymbolsWithoutUnderscore()
5189 void **p = symbolsWithoutUnderscore;
5190 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
5192 #undef SymI_NeedsProto
5193 #define SymI_NeedsProto(x) \
5194 __asm__ volatile(".long " # x);
5196 RTS_MACHO_NOUNDERLINE_SYMBOLS
5198 __asm__ volatile(".text");
5200 #undef SymI_NeedsProto
5201 #define SymI_NeedsProto(x) \
5202 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
5204 RTS_MACHO_NOUNDERLINE_SYMBOLS
5206 #undef SymI_NeedsProto
5212 * Figure out by how much to shift the entire Mach-O file in memory
5213 * when loading so that its single segment ends up 16-byte-aligned
5215 static int machoGetMisalignment( FILE * f )
5217 struct mach_header header;
5220 fread(&header, sizeof(header), 1, f);
5223 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
5224 if(header.magic != MH_MAGIC_64) {
5225 errorBelch("Bad magic. Expected: %08x, got: %08x.\n",
5226 MH_MAGIC_64, header->magic);
5230 if(header.magic != MH_MAGIC) {
5231 errorBelch("Bad magic. Expected: %08x, got: %08x.\n",
5232 MH_MAGIC, header->magic);
5237 misalignment = (header.sizeofcmds + sizeof(header))
5240 return misalignment ? (16 - misalignment) : 0;