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 #if defined(mingw32_HOST_OS)
37 // get protos for is*()
41 #ifdef HAVE_SYS_TYPES_H
42 #include <sys/types.h>
50 #ifdef HAVE_SYS_STAT_H
54 #if defined(HAVE_DLFCN_H)
58 #if defined(cygwin32_HOST_OS)
63 #ifdef HAVE_SYS_TIME_H
67 #include <sys/fcntl.h>
68 #include <sys/termios.h>
69 #include <sys/utime.h>
70 #include <sys/utsname.h>
74 #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)
85 #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)
86 # define OBJFORMAT_ELF
87 # include <regex.h> // regex is already used by dlopen() so this is OK
88 // to use here without requiring an additional lib
89 #elif defined(cygwin32_HOST_OS) || defined (mingw32_HOST_OS)
90 # define OBJFORMAT_PEi386
93 #elif defined(darwin_HOST_OS)
94 # define OBJFORMAT_MACHO
96 # include <mach-o/loader.h>
97 # include <mach-o/nlist.h>
98 # include <mach-o/reloc.h>
99 #if !defined(HAVE_DLFCN_H)
100 # include <mach-o/dyld.h>
102 #if defined(powerpc_HOST_ARCH)
103 # include <mach-o/ppc/reloc.h>
105 #if defined(x86_64_HOST_ARCH)
106 # include <mach-o/x86_64/reloc.h>
110 #if defined(x86_64_HOST_ARCH) && defined(darwin_HOST_OS)
114 /* Hash table mapping symbol names to Symbol */
115 static /*Str*/HashTable *symhash;
117 /* Hash table mapping symbol names to StgStablePtr */
118 static /*Str*/HashTable *stablehash;
120 /* List of currently loaded objects */
121 ObjectCode *objects = NULL; /* initially empty */
123 #if defined(OBJFORMAT_ELF)
124 static int ocVerifyImage_ELF ( ObjectCode* oc );
125 static int ocGetNames_ELF ( ObjectCode* oc );
126 static int ocResolve_ELF ( ObjectCode* oc );
127 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
128 static int ocAllocateSymbolExtras_ELF ( ObjectCode* oc );
130 #elif defined(OBJFORMAT_PEi386)
131 static int ocVerifyImage_PEi386 ( ObjectCode* oc );
132 static int ocGetNames_PEi386 ( ObjectCode* oc );
133 static int ocResolve_PEi386 ( ObjectCode* oc );
134 static void *lookupSymbolInDLLs ( unsigned char *lbl );
135 static void zapTrailingAtSign ( unsigned char *sym );
136 #elif defined(OBJFORMAT_MACHO)
137 static int ocVerifyImage_MachO ( ObjectCode* oc );
138 static int ocGetNames_MachO ( ObjectCode* oc );
139 static int ocResolve_MachO ( ObjectCode* oc );
142 static int machoGetMisalignment( FILE * );
144 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
145 static int ocAllocateSymbolExtras_MachO ( ObjectCode* oc );
147 #ifdef powerpc_HOST_ARCH
148 static void machoInitSymbolsWithoutUnderscore( void );
152 /* on x86_64 we have a problem with relocating symbol references in
153 * code that was compiled without -fPIC. By default, the small memory
154 * model is used, which assumes that symbol references can fit in a
155 * 32-bit slot. The system dynamic linker makes this work for
156 * references to shared libraries by either (a) allocating a jump
157 * table slot for code references, or (b) moving the symbol at load
158 * time (and copying its contents, if necessary) for data references.
160 * We unfortunately can't tell whether symbol references are to code
161 * or data. So for now we assume they are code (the vast majority
162 * are), and allocate jump-table slots. Unfortunately this will
163 * SILENTLY generate crashing code for data references. This hack is
164 * enabled by X86_64_ELF_NONPIC_HACK.
166 * One workaround is to use shared Haskell libraries. This is
167 * coming. Another workaround is to keep the static libraries but
168 * compile them with -fPIC, because that will generate PIC references
169 * to data which can be relocated. The PIC code is still too green to
170 * do this systematically, though.
173 * See thread http://www.haskell.org/pipermail/cvs-ghc/2007-September/038458.html
175 * Naming Scheme for Symbol Macros
177 * SymI_*: symbol is internal to the RTS. It resides in an object
178 * file/library that is statically.
179 * SymE_*: symbol is external to the RTS library. It might be linked
182 * Sym*_HasProto : the symbol prototype is imported in an include file
183 * or defined explicitly
184 * Sym*_NeedsProto: the symbol is undefined and we add a dummy
185 * default proto extern void sym(void);
187 #define X86_64_ELF_NONPIC_HACK 1
189 /* Link objects into the lower 2Gb on x86_64. GHC assumes the
190 * small memory model on this architecture (see gcc docs,
193 * MAP_32BIT not available on OpenBSD/amd64
195 #if defined(x86_64_HOST_ARCH) && defined(MAP_32BIT)
196 #define TRY_MAP_32BIT MAP_32BIT
198 #define TRY_MAP_32BIT 0
202 * Due to the small memory model (see above), on x86_64 we have to map
203 * all our non-PIC object files into the low 2Gb of the address space
204 * (why 2Gb and not 4Gb? Because all addresses must be reachable
205 * using a 32-bit signed PC-relative offset). On Linux we can do this
206 * using the MAP_32BIT flag to mmap(), however on other OSs
207 * (e.g. *BSD, see #2063, and also on Linux inside Xen, see #2512), we
208 * can't do this. So on these systems, we have to pick a base address
209 * in the low 2Gb of the address space and try to allocate memory from
212 * We pick a default address based on the OS, but also make this
213 * configurable via an RTS flag (+RTS -xm)
215 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
217 #if defined(MAP_32BIT)
218 // Try to use MAP_32BIT
219 #define MMAP_32BIT_BASE_DEFAULT 0
222 #define MMAP_32BIT_BASE_DEFAULT 0x40000000
225 static void *mmap_32bit_base = (void *)MMAP_32BIT_BASE_DEFAULT;
228 /* MAP_ANONYMOUS is MAP_ANON on some systems, e.g. OpenBSD */
229 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
230 #define MAP_ANONYMOUS MAP_ANON
233 /* -----------------------------------------------------------------------------
234 * Built-in symbols from the RTS
237 typedef struct _RtsSymbolVal {
242 #define Maybe_Stable_Names SymI_HasProto(stg_mkWeakzh) \
243 SymI_HasProto(stg_mkWeakForeignEnvzh) \
244 SymI_HasProto(stg_makeStableNamezh) \
245 SymI_HasProto(stg_finalizzeWeakzh)
247 #if !defined (mingw32_HOST_OS)
248 #define RTS_POSIX_ONLY_SYMBOLS \
249 SymI_HasProto(__hscore_get_saved_termios) \
250 SymI_HasProto(__hscore_set_saved_termios) \
251 SymI_HasProto(shutdownHaskellAndSignal) \
252 SymI_HasProto(lockFile) \
253 SymI_HasProto(unlockFile) \
254 SymI_HasProto(signal_handlers) \
255 SymI_HasProto(stg_sig_install) \
256 SymI_NeedsProto(nocldstop)
259 #if defined (cygwin32_HOST_OS)
260 #define RTS_MINGW_ONLY_SYMBOLS /**/
261 /* Don't have the ability to read import libs / archives, so
262 * we have to stupidly list a lot of what libcygwin.a
265 #define RTS_CYGWIN_ONLY_SYMBOLS \
266 SymI_HasProto(regfree) \
267 SymI_HasProto(regexec) \
268 SymI_HasProto(regerror) \
269 SymI_HasProto(regcomp) \
270 SymI_HasProto(__errno) \
271 SymI_HasProto(access) \
272 SymI_HasProto(chmod) \
273 SymI_HasProto(chdir) \
274 SymI_HasProto(close) \
275 SymI_HasProto(creat) \
277 SymI_HasProto(dup2) \
278 SymI_HasProto(fstat) \
279 SymI_HasProto(fcntl) \
280 SymI_HasProto(getcwd) \
281 SymI_HasProto(getenv) \
282 SymI_HasProto(lseek) \
283 SymI_HasProto(open) \
284 SymI_HasProto(fpathconf) \
285 SymI_HasProto(pathconf) \
286 SymI_HasProto(stat) \
288 SymI_HasProto(tanh) \
289 SymI_HasProto(cosh) \
290 SymI_HasProto(sinh) \
291 SymI_HasProto(atan) \
292 SymI_HasProto(acos) \
293 SymI_HasProto(asin) \
299 SymI_HasProto(sqrt) \
300 SymI_HasProto(localtime_r) \
301 SymI_HasProto(gmtime_r) \
302 SymI_HasProto(mktime) \
303 SymI_NeedsProto(_imp___tzname) \
304 SymI_HasProto(gettimeofday) \
305 SymI_HasProto(timezone) \
306 SymI_HasProto(tcgetattr) \
307 SymI_HasProto(tcsetattr) \
308 SymI_HasProto(memcpy) \
309 SymI_HasProto(memmove) \
310 SymI_HasProto(realloc) \
311 SymI_HasProto(malloc) \
312 SymI_HasProto(free) \
313 SymI_HasProto(fork) \
314 SymI_HasProto(lstat) \
315 SymI_HasProto(isatty) \
316 SymI_HasProto(mkdir) \
317 SymI_HasProto(opendir) \
318 SymI_HasProto(readdir) \
319 SymI_HasProto(rewinddir) \
320 SymI_HasProto(closedir) \
321 SymI_HasProto(link) \
322 SymI_HasProto(mkfifo) \
323 SymI_HasProto(pipe) \
324 SymI_HasProto(read) \
325 SymI_HasProto(rename) \
326 SymI_HasProto(rmdir) \
327 SymI_HasProto(select) \
328 SymI_HasProto(system) \
329 SymI_HasProto(write) \
330 SymI_HasProto(strcmp) \
331 SymI_HasProto(strcpy) \
332 SymI_HasProto(strncpy) \
333 SymI_HasProto(strerror) \
334 SymI_HasProto(sigaddset) \
335 SymI_HasProto(sigemptyset) \
336 SymI_HasProto(sigprocmask) \
337 SymI_HasProto(umask) \
338 SymI_HasProto(uname) \
339 SymI_HasProto(unlink) \
340 SymI_HasProto(utime) \
341 SymI_HasProto(waitpid)
343 #elif !defined(mingw32_HOST_OS)
344 #define RTS_MINGW_ONLY_SYMBOLS /**/
345 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
346 #else /* defined(mingw32_HOST_OS) */
347 #define RTS_POSIX_ONLY_SYMBOLS /**/
348 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
350 /* Extra syms gen'ed by mingw-2's gcc-3.2: */
352 #define RTS_MINGW_EXTRA_SYMS \
353 SymI_NeedsProto(_imp____mb_cur_max) \
354 SymI_NeedsProto(_imp___pctype)
356 #define RTS_MINGW_EXTRA_SYMS
359 #if HAVE_GETTIMEOFDAY
360 #define RTS_MINGW_GETTIMEOFDAY_SYM SymI_NeedsProto(gettimeofday)
362 #define RTS_MINGW_GETTIMEOFDAY_SYM /**/
365 #if HAVE___MINGW_VFPRINTF
366 #define RTS___MINGW_VFPRINTF_SYM SymI_HasProto(__mingw_vfprintf)
368 #define RTS___MINGW_VFPRINTF_SYM /**/
371 /* These are statically linked from the mingw libraries into the ghc
372 executable, so we have to employ this hack. */
373 #define RTS_MINGW_ONLY_SYMBOLS \
374 SymI_HasProto(stg_asyncReadzh) \
375 SymI_HasProto(stg_asyncWritezh) \
376 SymI_HasProto(stg_asyncDoProczh) \
377 SymI_HasProto(memset) \
378 SymI_HasProto(inet_ntoa) \
379 SymI_HasProto(inet_addr) \
380 SymI_HasProto(htonl) \
381 SymI_HasProto(recvfrom) \
382 SymI_HasProto(listen) \
383 SymI_HasProto(bind) \
384 SymI_HasProto(shutdown) \
385 SymI_HasProto(connect) \
386 SymI_HasProto(htons) \
387 SymI_HasProto(ntohs) \
388 SymI_HasProto(getservbyname) \
389 SymI_HasProto(getservbyport) \
390 SymI_HasProto(getprotobynumber) \
391 SymI_HasProto(getprotobyname) \
392 SymI_HasProto(gethostbyname) \
393 SymI_HasProto(gethostbyaddr) \
394 SymI_HasProto(gethostname) \
395 SymI_HasProto(strcpy) \
396 SymI_HasProto(strncpy) \
397 SymI_HasProto(abort) \
398 SymI_NeedsProto(_alloca) \
399 SymI_HasProto(isxdigit) \
400 SymI_HasProto(isupper) \
401 SymI_HasProto(ispunct) \
402 SymI_HasProto(islower) \
403 SymI_HasProto(isspace) \
404 SymI_HasProto(isprint) \
405 SymI_HasProto(isdigit) \
406 SymI_HasProto(iscntrl) \
407 SymI_HasProto(isalpha) \
408 SymI_HasProto(isalnum) \
409 SymI_HasProto(isascii) \
410 RTS___MINGW_VFPRINTF_SYM \
411 SymI_HasProto(strcmp) \
412 SymI_HasProto(memmove) \
413 SymI_HasProto(realloc) \
414 SymI_HasProto(malloc) \
416 SymI_HasProto(tanh) \
417 SymI_HasProto(cosh) \
418 SymI_HasProto(sinh) \
419 SymI_HasProto(atan) \
420 SymI_HasProto(acos) \
421 SymI_HasProto(asin) \
427 SymI_HasProto(sqrt) \
428 SymI_HasProto(powf) \
429 SymI_HasProto(tanhf) \
430 SymI_HasProto(coshf) \
431 SymI_HasProto(sinhf) \
432 SymI_HasProto(atanf) \
433 SymI_HasProto(acosf) \
434 SymI_HasProto(asinf) \
435 SymI_HasProto(tanf) \
436 SymI_HasProto(cosf) \
437 SymI_HasProto(sinf) \
438 SymI_HasProto(expf) \
439 SymI_HasProto(logf) \
440 SymI_HasProto(sqrtf) \
442 SymI_HasProto(erfc) \
443 SymI_HasProto(erff) \
444 SymI_HasProto(erfcf) \
445 SymI_HasProto(memcpy) \
446 SymI_HasProto(rts_InstallConsoleEvent) \
447 SymI_HasProto(rts_ConsoleHandlerDone) \
448 SymI_NeedsProto(mktime) \
449 SymI_NeedsProto(_imp___timezone) \
450 SymI_NeedsProto(_imp___tzname) \
451 SymI_NeedsProto(_imp__tzname) \
452 SymI_NeedsProto(_imp___iob) \
453 SymI_NeedsProto(_imp___osver) \
454 SymI_NeedsProto(localtime) \
455 SymI_NeedsProto(gmtime) \
456 SymI_NeedsProto(opendir) \
457 SymI_NeedsProto(readdir) \
458 SymI_NeedsProto(rewinddir) \
459 RTS_MINGW_EXTRA_SYMS \
460 RTS_MINGW_GETTIMEOFDAY_SYM \
461 SymI_NeedsProto(closedir)
464 #if defined(darwin_HOST_OS) && HAVE_PRINTF_LDBLSTUB
465 #define RTS_DARWIN_ONLY_SYMBOLS \
466 SymI_NeedsProto(asprintf$LDBLStub) \
467 SymI_NeedsProto(err$LDBLStub) \
468 SymI_NeedsProto(errc$LDBLStub) \
469 SymI_NeedsProto(errx$LDBLStub) \
470 SymI_NeedsProto(fprintf$LDBLStub) \
471 SymI_NeedsProto(fscanf$LDBLStub) \
472 SymI_NeedsProto(fwprintf$LDBLStub) \
473 SymI_NeedsProto(fwscanf$LDBLStub) \
474 SymI_NeedsProto(printf$LDBLStub) \
475 SymI_NeedsProto(scanf$LDBLStub) \
476 SymI_NeedsProto(snprintf$LDBLStub) \
477 SymI_NeedsProto(sprintf$LDBLStub) \
478 SymI_NeedsProto(sscanf$LDBLStub) \
479 SymI_NeedsProto(strtold$LDBLStub) \
480 SymI_NeedsProto(swprintf$LDBLStub) \
481 SymI_NeedsProto(swscanf$LDBLStub) \
482 SymI_NeedsProto(syslog$LDBLStub) \
483 SymI_NeedsProto(vasprintf$LDBLStub) \
484 SymI_NeedsProto(verr$LDBLStub) \
485 SymI_NeedsProto(verrc$LDBLStub) \
486 SymI_NeedsProto(verrx$LDBLStub) \
487 SymI_NeedsProto(vfprintf$LDBLStub) \
488 SymI_NeedsProto(vfscanf$LDBLStub) \
489 SymI_NeedsProto(vfwprintf$LDBLStub) \
490 SymI_NeedsProto(vfwscanf$LDBLStub) \
491 SymI_NeedsProto(vprintf$LDBLStub) \
492 SymI_NeedsProto(vscanf$LDBLStub) \
493 SymI_NeedsProto(vsnprintf$LDBLStub) \
494 SymI_NeedsProto(vsprintf$LDBLStub) \
495 SymI_NeedsProto(vsscanf$LDBLStub) \
496 SymI_NeedsProto(vswprintf$LDBLStub) \
497 SymI_NeedsProto(vswscanf$LDBLStub) \
498 SymI_NeedsProto(vsyslog$LDBLStub) \
499 SymI_NeedsProto(vwarn$LDBLStub) \
500 SymI_NeedsProto(vwarnc$LDBLStub) \
501 SymI_NeedsProto(vwarnx$LDBLStub) \
502 SymI_NeedsProto(vwprintf$LDBLStub) \
503 SymI_NeedsProto(vwscanf$LDBLStub) \
504 SymI_NeedsProto(warn$LDBLStub) \
505 SymI_NeedsProto(warnc$LDBLStub) \
506 SymI_NeedsProto(warnx$LDBLStub) \
507 SymI_NeedsProto(wcstold$LDBLStub) \
508 SymI_NeedsProto(wprintf$LDBLStub) \
509 SymI_NeedsProto(wscanf$LDBLStub)
511 #define RTS_DARWIN_ONLY_SYMBOLS
515 # define MAIN_CAP_SYM SymI_HasProto(MainCapability)
517 # define MAIN_CAP_SYM
520 #if !defined(mingw32_HOST_OS)
521 #define RTS_USER_SIGNALS_SYMBOLS \
522 SymI_HasProto(setIOManagerControlFd) \
523 SymI_HasProto(setIOManagerWakeupFd) \
524 SymI_HasProto(ioManagerWakeup) \
525 SymI_HasProto(blockUserSignals) \
526 SymI_HasProto(unblockUserSignals)
528 #define RTS_USER_SIGNALS_SYMBOLS \
529 SymI_HasProto(ioManagerWakeup) \
530 SymI_HasProto(sendIOManagerEvent) \
531 SymI_HasProto(readIOManagerEvent) \
532 SymI_HasProto(getIOManagerEvent) \
533 SymI_HasProto(console_handler)
536 #define RTS_LIBFFI_SYMBOLS \
537 SymE_NeedsProto(ffi_prep_cif) \
538 SymE_NeedsProto(ffi_call) \
539 SymE_NeedsProto(ffi_type_void) \
540 SymE_NeedsProto(ffi_type_float) \
541 SymE_NeedsProto(ffi_type_double) \
542 SymE_NeedsProto(ffi_type_sint64) \
543 SymE_NeedsProto(ffi_type_uint64) \
544 SymE_NeedsProto(ffi_type_sint32) \
545 SymE_NeedsProto(ffi_type_uint32) \
546 SymE_NeedsProto(ffi_type_sint16) \
547 SymE_NeedsProto(ffi_type_uint16) \
548 SymE_NeedsProto(ffi_type_sint8) \
549 SymE_NeedsProto(ffi_type_uint8) \
550 SymE_NeedsProto(ffi_type_pointer)
552 #ifdef TABLES_NEXT_TO_CODE
553 #define RTS_RET_SYMBOLS /* nothing */
555 #define RTS_RET_SYMBOLS \
556 SymI_HasProto(stg_enter_ret) \
557 SymI_HasProto(stg_gc_fun_ret) \
558 SymI_HasProto(stg_ap_v_ret) \
559 SymI_HasProto(stg_ap_f_ret) \
560 SymI_HasProto(stg_ap_d_ret) \
561 SymI_HasProto(stg_ap_l_ret) \
562 SymI_HasProto(stg_ap_n_ret) \
563 SymI_HasProto(stg_ap_p_ret) \
564 SymI_HasProto(stg_ap_pv_ret) \
565 SymI_HasProto(stg_ap_pp_ret) \
566 SymI_HasProto(stg_ap_ppv_ret) \
567 SymI_HasProto(stg_ap_ppp_ret) \
568 SymI_HasProto(stg_ap_pppv_ret) \
569 SymI_HasProto(stg_ap_pppp_ret) \
570 SymI_HasProto(stg_ap_ppppp_ret) \
571 SymI_HasProto(stg_ap_pppppp_ret)
574 /* Modules compiled with -ticky may mention ticky counters */
575 /* This list should marry up with the one in $(TOP)/includes/stg/Ticky.h */
576 #define RTS_TICKY_SYMBOLS \
577 SymI_NeedsProto(ticky_entry_ctrs) \
578 SymI_NeedsProto(top_ct) \
580 SymI_HasProto(ENT_VIA_NODE_ctr) \
581 SymI_HasProto(ENT_STATIC_THK_ctr) \
582 SymI_HasProto(ENT_DYN_THK_ctr) \
583 SymI_HasProto(ENT_STATIC_FUN_DIRECT_ctr) \
584 SymI_HasProto(ENT_DYN_FUN_DIRECT_ctr) \
585 SymI_HasProto(ENT_STATIC_CON_ctr) \
586 SymI_HasProto(ENT_DYN_CON_ctr) \
587 SymI_HasProto(ENT_STATIC_IND_ctr) \
588 SymI_HasProto(ENT_DYN_IND_ctr) \
589 SymI_HasProto(ENT_PERM_IND_ctr) \
590 SymI_HasProto(ENT_PAP_ctr) \
591 SymI_HasProto(ENT_AP_ctr) \
592 SymI_HasProto(ENT_AP_STACK_ctr) \
593 SymI_HasProto(ENT_BH_ctr) \
594 SymI_HasProto(UNKNOWN_CALL_ctr) \
595 SymI_HasProto(SLOW_CALL_v_ctr) \
596 SymI_HasProto(SLOW_CALL_f_ctr) \
597 SymI_HasProto(SLOW_CALL_d_ctr) \
598 SymI_HasProto(SLOW_CALL_l_ctr) \
599 SymI_HasProto(SLOW_CALL_n_ctr) \
600 SymI_HasProto(SLOW_CALL_p_ctr) \
601 SymI_HasProto(SLOW_CALL_pv_ctr) \
602 SymI_HasProto(SLOW_CALL_pp_ctr) \
603 SymI_HasProto(SLOW_CALL_ppv_ctr) \
604 SymI_HasProto(SLOW_CALL_ppp_ctr) \
605 SymI_HasProto(SLOW_CALL_pppv_ctr) \
606 SymI_HasProto(SLOW_CALL_pppp_ctr) \
607 SymI_HasProto(SLOW_CALL_ppppp_ctr) \
608 SymI_HasProto(SLOW_CALL_pppppp_ctr) \
609 SymI_HasProto(SLOW_CALL_OTHER_ctr) \
610 SymI_HasProto(ticky_slow_call_unevald) \
611 SymI_HasProto(SLOW_CALL_ctr) \
612 SymI_HasProto(MULTI_CHUNK_SLOW_CALL_ctr) \
613 SymI_HasProto(MULTI_CHUNK_SLOW_CALL_CHUNKS_ctr) \
614 SymI_HasProto(KNOWN_CALL_ctr) \
615 SymI_HasProto(KNOWN_CALL_TOO_FEW_ARGS_ctr) \
616 SymI_HasProto(KNOWN_CALL_EXTRA_ARGS_ctr) \
617 SymI_HasProto(SLOW_CALL_FUN_TOO_FEW_ctr) \
618 SymI_HasProto(SLOW_CALL_FUN_CORRECT_ctr) \
619 SymI_HasProto(SLOW_CALL_FUN_TOO_MANY_ctr) \
620 SymI_HasProto(SLOW_CALL_PAP_TOO_FEW_ctr) \
621 SymI_HasProto(SLOW_CALL_PAP_CORRECT_ctr) \
622 SymI_HasProto(SLOW_CALL_PAP_TOO_MANY_ctr) \
623 SymI_HasProto(SLOW_CALL_UNEVALD_ctr) \
624 SymI_HasProto(UPDF_OMITTED_ctr) \
625 SymI_HasProto(UPDF_PUSHED_ctr) \
626 SymI_HasProto(CATCHF_PUSHED_ctr) \
627 SymI_HasProto(UPDF_RCC_PUSHED_ctr) \
628 SymI_HasProto(UPDF_RCC_OMITTED_ctr) \
629 SymI_HasProto(UPD_SQUEEZED_ctr) \
630 SymI_HasProto(UPD_CON_IN_NEW_ctr) \
631 SymI_HasProto(UPD_CON_IN_PLACE_ctr) \
632 SymI_HasProto(UPD_PAP_IN_NEW_ctr) \
633 SymI_HasProto(UPD_PAP_IN_PLACE_ctr) \
634 SymI_HasProto(ALLOC_HEAP_ctr) \
635 SymI_HasProto(ALLOC_HEAP_tot) \
636 SymI_HasProto(ALLOC_FUN_ctr) \
637 SymI_HasProto(ALLOC_FUN_adm) \
638 SymI_HasProto(ALLOC_FUN_gds) \
639 SymI_HasProto(ALLOC_FUN_slp) \
640 SymI_HasProto(UPD_NEW_IND_ctr) \
641 SymI_HasProto(UPD_NEW_PERM_IND_ctr) \
642 SymI_HasProto(UPD_OLD_IND_ctr) \
643 SymI_HasProto(UPD_OLD_PERM_IND_ctr) \
644 SymI_HasProto(UPD_BH_UPDATABLE_ctr) \
645 SymI_HasProto(UPD_BH_SINGLE_ENTRY_ctr) \
646 SymI_HasProto(UPD_CAF_BH_UPDATABLE_ctr) \
647 SymI_HasProto(UPD_CAF_BH_SINGLE_ENTRY_ctr) \
648 SymI_HasProto(GC_SEL_ABANDONED_ctr) \
649 SymI_HasProto(GC_SEL_MINOR_ctr) \
650 SymI_HasProto(GC_SEL_MAJOR_ctr) \
651 SymI_HasProto(GC_FAILED_PROMOTION_ctr) \
652 SymI_HasProto(ALLOC_UP_THK_ctr) \
653 SymI_HasProto(ALLOC_SE_THK_ctr) \
654 SymI_HasProto(ALLOC_THK_adm) \
655 SymI_HasProto(ALLOC_THK_gds) \
656 SymI_HasProto(ALLOC_THK_slp) \
657 SymI_HasProto(ALLOC_CON_ctr) \
658 SymI_HasProto(ALLOC_CON_adm) \
659 SymI_HasProto(ALLOC_CON_gds) \
660 SymI_HasProto(ALLOC_CON_slp) \
661 SymI_HasProto(ALLOC_TUP_ctr) \
662 SymI_HasProto(ALLOC_TUP_adm) \
663 SymI_HasProto(ALLOC_TUP_gds) \
664 SymI_HasProto(ALLOC_TUP_slp) \
665 SymI_HasProto(ALLOC_BH_ctr) \
666 SymI_HasProto(ALLOC_BH_adm) \
667 SymI_HasProto(ALLOC_BH_gds) \
668 SymI_HasProto(ALLOC_BH_slp) \
669 SymI_HasProto(ALLOC_PRIM_ctr) \
670 SymI_HasProto(ALLOC_PRIM_adm) \
671 SymI_HasProto(ALLOC_PRIM_gds) \
672 SymI_HasProto(ALLOC_PRIM_slp) \
673 SymI_HasProto(ALLOC_PAP_ctr) \
674 SymI_HasProto(ALLOC_PAP_adm) \
675 SymI_HasProto(ALLOC_PAP_gds) \
676 SymI_HasProto(ALLOC_PAP_slp) \
677 SymI_HasProto(ALLOC_TSO_ctr) \
678 SymI_HasProto(ALLOC_TSO_adm) \
679 SymI_HasProto(ALLOC_TSO_gds) \
680 SymI_HasProto(ALLOC_TSO_slp) \
681 SymI_HasProto(RET_NEW_ctr) \
682 SymI_HasProto(RET_OLD_ctr) \
683 SymI_HasProto(RET_UNBOXED_TUP_ctr) \
684 SymI_HasProto(RET_SEMI_loads_avoided)
687 // On most platforms, the garbage collector rewrites references
688 // to small integer and char objects to a set of common, shared ones.
690 // We don't do this when compiling to Windows DLLs at the moment because
691 // it doesn't support cross package data references well.
693 #if defined(__PIC__) && defined(mingw32_HOST_OS)
694 #define RTS_INTCHAR_SYMBOLS
696 #define RTS_INTCHAR_SYMBOLS \
697 SymI_HasProto(stg_CHARLIKE_closure) \
698 SymI_HasProto(stg_INTLIKE_closure)
702 #define RTS_SYMBOLS \
705 SymI_HasProto(StgReturn) \
706 SymI_HasProto(stg_enter_info) \
707 SymI_HasProto(stg_gc_void_info) \
708 SymI_HasProto(__stg_gc_enter_1) \
709 SymI_HasProto(stg_gc_noregs) \
710 SymI_HasProto(stg_gc_unpt_r1_info) \
711 SymI_HasProto(stg_gc_unpt_r1) \
712 SymI_HasProto(stg_gc_unbx_r1_info) \
713 SymI_HasProto(stg_gc_unbx_r1) \
714 SymI_HasProto(stg_gc_f1_info) \
715 SymI_HasProto(stg_gc_f1) \
716 SymI_HasProto(stg_gc_d1_info) \
717 SymI_HasProto(stg_gc_d1) \
718 SymI_HasProto(stg_gc_l1_info) \
719 SymI_HasProto(stg_gc_l1) \
720 SymI_HasProto(__stg_gc_fun) \
721 SymI_HasProto(stg_gc_fun_info) \
722 SymI_HasProto(stg_gc_gen) \
723 SymI_HasProto(stg_gc_gen_info) \
724 SymI_HasProto(stg_gc_gen_hp) \
725 SymI_HasProto(stg_gc_ut) \
726 SymI_HasProto(stg_gen_yield) \
727 SymI_HasProto(stg_yield_noregs) \
728 SymI_HasProto(stg_yield_to_interpreter) \
729 SymI_HasProto(stg_gen_block) \
730 SymI_HasProto(stg_block_noregs) \
731 SymI_HasProto(stg_block_1) \
732 SymI_HasProto(stg_block_takemvar) \
733 SymI_HasProto(stg_block_putmvar) \
735 SymI_HasProto(MallocFailHook) \
736 SymI_HasProto(OnExitHook) \
737 SymI_HasProto(OutOfHeapHook) \
738 SymI_HasProto(StackOverflowHook) \
739 SymI_HasProto(addDLL) \
740 SymI_HasProto(__int_encodeDouble) \
741 SymI_HasProto(__word_encodeDouble) \
742 SymI_HasProto(__2Int_encodeDouble) \
743 SymI_HasProto(__int_encodeFloat) \
744 SymI_HasProto(__word_encodeFloat) \
745 SymI_HasProto(stg_atomicallyzh) \
746 SymI_HasProto(barf) \
747 SymI_HasProto(debugBelch) \
748 SymI_HasProto(errorBelch) \
749 SymI_HasProto(sysErrorBelch) \
750 SymI_HasProto(stg_getMaskingStatezh) \
751 SymI_HasProto(stg_maskAsyncExceptionszh) \
752 SymI_HasProto(stg_maskUninterruptiblezh) \
753 SymI_HasProto(stg_catchzh) \
754 SymI_HasProto(stg_catchRetryzh) \
755 SymI_HasProto(stg_catchSTMzh) \
756 SymI_HasProto(stg_checkzh) \
757 SymI_HasProto(closure_flags) \
758 SymI_HasProto(cmp_thread) \
759 SymI_HasProto(createAdjustor) \
760 SymI_HasProto(stg_decodeDoublezu2Intzh) \
761 SymI_HasProto(stg_decodeFloatzuIntzh) \
762 SymI_HasProto(defaultsHook) \
763 SymI_HasProto(stg_delayzh) \
764 SymI_HasProto(stg_deRefWeakzh) \
765 SymI_HasProto(stg_deRefStablePtrzh) \
766 SymI_HasProto(dirty_MUT_VAR) \
767 SymI_HasProto(stg_forkzh) \
768 SymI_HasProto(stg_forkOnzh) \
769 SymI_HasProto(forkProcess) \
770 SymI_HasProto(forkOS_createThread) \
771 SymI_HasProto(freeHaskellFunctionPtr) \
772 SymI_HasProto(getOrSetTypeableStore) \
773 SymI_HasProto(getOrSetGHCConcSignalSignalHandlerStore) \
774 SymI_HasProto(getOrSetGHCConcWindowsPendingDelaysStore) \
775 SymI_HasProto(getOrSetGHCConcWindowsIOManagerThreadStore) \
776 SymI_HasProto(getOrSetGHCConcWindowsProddingStore) \
777 SymI_HasProto(getOrSetSystemEventThreadEventManagerStore) \
778 SymI_HasProto(getOrSetSystemEventThreadIOManagerThreadStore) \
779 SymI_HasProto(genSymZh) \
780 SymI_HasProto(genericRaise) \
781 SymI_HasProto(getProgArgv) \
782 SymI_HasProto(getFullProgArgv) \
783 SymI_HasProto(getStablePtr) \
784 SymI_HasProto(hs_init) \
785 SymI_HasProto(hs_exit) \
786 SymI_HasProto(hs_set_argv) \
787 SymI_HasProto(hs_add_root) \
788 SymI_HasProto(hs_perform_gc) \
789 SymI_HasProto(hs_free_stable_ptr) \
790 SymI_HasProto(hs_free_fun_ptr) \
791 SymI_HasProto(hs_hpc_rootModule) \
792 SymI_HasProto(hs_hpc_module) \
793 SymI_HasProto(initLinker) \
794 SymI_HasProto(stg_unpackClosurezh) \
795 SymI_HasProto(stg_getApStackValzh) \
796 SymI_HasProto(stg_getSparkzh) \
797 SymI_HasProto(stg_numSparkszh) \
798 SymI_HasProto(stg_isCurrentThreadBoundzh) \
799 SymI_HasProto(stg_isEmptyMVarzh) \
800 SymI_HasProto(stg_killThreadzh) \
801 SymI_HasProto(loadObj) \
802 SymI_HasProto(insertStableSymbol) \
803 SymI_HasProto(insertSymbol) \
804 SymI_HasProto(lookupSymbol) \
805 SymI_HasProto(stg_makeStablePtrzh) \
806 SymI_HasProto(stg_mkApUpd0zh) \
807 SymI_HasProto(stg_myThreadIdzh) \
808 SymI_HasProto(stg_labelThreadzh) \
809 SymI_HasProto(stg_newArrayzh) \
810 SymI_HasProto(stg_newBCOzh) \
811 SymI_HasProto(stg_newByteArrayzh) \
812 SymI_HasProto_redirect(newCAF, newDynCAF) \
813 SymI_HasProto(stg_newMVarzh) \
814 SymI_HasProto(stg_newMutVarzh) \
815 SymI_HasProto(stg_newTVarzh) \
816 SymI_HasProto(stg_noDuplicatezh) \
817 SymI_HasProto(stg_atomicModifyMutVarzh) \
818 SymI_HasProto(stg_newPinnedByteArrayzh) \
819 SymI_HasProto(stg_newAlignedPinnedByteArrayzh) \
820 SymI_HasProto(newSpark) \
821 SymI_HasProto(performGC) \
822 SymI_HasProto(performMajorGC) \
823 SymI_HasProto(prog_argc) \
824 SymI_HasProto(prog_argv) \
825 SymI_HasProto(stg_putMVarzh) \
826 SymI_HasProto(stg_raisezh) \
827 SymI_HasProto(stg_raiseIOzh) \
828 SymI_HasProto(stg_readTVarzh) \
829 SymI_HasProto(stg_readTVarIOzh) \
830 SymI_HasProto(resumeThread) \
831 SymI_HasProto(resolveObjs) \
832 SymI_HasProto(stg_retryzh) \
833 SymI_HasProto(rts_apply) \
834 SymI_HasProto(rts_checkSchedStatus) \
835 SymI_HasProto(rts_eval) \
836 SymI_HasProto(rts_evalIO) \
837 SymI_HasProto(rts_evalLazyIO) \
838 SymI_HasProto(rts_evalStableIO) \
839 SymI_HasProto(rts_eval_) \
840 SymI_HasProto(rts_getBool) \
841 SymI_HasProto(rts_getChar) \
842 SymI_HasProto(rts_getDouble) \
843 SymI_HasProto(rts_getFloat) \
844 SymI_HasProto(rts_getInt) \
845 SymI_HasProto(rts_getInt8) \
846 SymI_HasProto(rts_getInt16) \
847 SymI_HasProto(rts_getInt32) \
848 SymI_HasProto(rts_getInt64) \
849 SymI_HasProto(rts_getPtr) \
850 SymI_HasProto(rts_getFunPtr) \
851 SymI_HasProto(rts_getStablePtr) \
852 SymI_HasProto(rts_getThreadId) \
853 SymI_HasProto(rts_getWord) \
854 SymI_HasProto(rts_getWord8) \
855 SymI_HasProto(rts_getWord16) \
856 SymI_HasProto(rts_getWord32) \
857 SymI_HasProto(rts_getWord64) \
858 SymI_HasProto(rts_lock) \
859 SymI_HasProto(rts_mkBool) \
860 SymI_HasProto(rts_mkChar) \
861 SymI_HasProto(rts_mkDouble) \
862 SymI_HasProto(rts_mkFloat) \
863 SymI_HasProto(rts_mkInt) \
864 SymI_HasProto(rts_mkInt8) \
865 SymI_HasProto(rts_mkInt16) \
866 SymI_HasProto(rts_mkInt32) \
867 SymI_HasProto(rts_mkInt64) \
868 SymI_HasProto(rts_mkPtr) \
869 SymI_HasProto(rts_mkFunPtr) \
870 SymI_HasProto(rts_mkStablePtr) \
871 SymI_HasProto(rts_mkString) \
872 SymI_HasProto(rts_mkWord) \
873 SymI_HasProto(rts_mkWord8) \
874 SymI_HasProto(rts_mkWord16) \
875 SymI_HasProto(rts_mkWord32) \
876 SymI_HasProto(rts_mkWord64) \
877 SymI_HasProto(rts_unlock) \
878 SymI_HasProto(rts_unsafeGetMyCapability) \
879 SymI_HasProto(rtsSupportsBoundThreads) \
880 SymI_HasProto(rts_isProfiled) \
881 SymI_HasProto(setProgArgv) \
882 SymI_HasProto(startupHaskell) \
883 SymI_HasProto(shutdownHaskell) \
884 SymI_HasProto(shutdownHaskellAndExit) \
885 SymI_HasProto(stable_ptr_table) \
886 SymI_HasProto(stackOverflow) \
887 SymI_HasProto(stg_CAF_BLACKHOLE_info) \
888 SymI_HasProto(stg_BLACKHOLE_info) \
889 SymI_HasProto(__stg_EAGER_BLACKHOLE_info) \
890 SymI_HasProto(stg_BLOCKING_QUEUE_CLEAN_info) \
891 SymI_HasProto(stg_BLOCKING_QUEUE_DIRTY_info) \
892 SymI_HasProto(startTimer) \
893 SymI_HasProto(stg_MVAR_CLEAN_info) \
894 SymI_HasProto(stg_MVAR_DIRTY_info) \
895 SymI_HasProto(stg_IND_STATIC_info) \
896 SymI_HasProto(stg_ARR_WORDS_info) \
897 SymI_HasProto(stg_MUT_ARR_PTRS_DIRTY_info) \
898 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN_info) \
899 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN0_info) \
900 SymI_HasProto(stg_WEAK_info) \
901 SymI_HasProto(stg_ap_v_info) \
902 SymI_HasProto(stg_ap_f_info) \
903 SymI_HasProto(stg_ap_d_info) \
904 SymI_HasProto(stg_ap_l_info) \
905 SymI_HasProto(stg_ap_n_info) \
906 SymI_HasProto(stg_ap_p_info) \
907 SymI_HasProto(stg_ap_pv_info) \
908 SymI_HasProto(stg_ap_pp_info) \
909 SymI_HasProto(stg_ap_ppv_info) \
910 SymI_HasProto(stg_ap_ppp_info) \
911 SymI_HasProto(stg_ap_pppv_info) \
912 SymI_HasProto(stg_ap_pppp_info) \
913 SymI_HasProto(stg_ap_ppppp_info) \
914 SymI_HasProto(stg_ap_pppppp_info) \
915 SymI_HasProto(stg_ap_0_fast) \
916 SymI_HasProto(stg_ap_v_fast) \
917 SymI_HasProto(stg_ap_f_fast) \
918 SymI_HasProto(stg_ap_d_fast) \
919 SymI_HasProto(stg_ap_l_fast) \
920 SymI_HasProto(stg_ap_n_fast) \
921 SymI_HasProto(stg_ap_p_fast) \
922 SymI_HasProto(stg_ap_pv_fast) \
923 SymI_HasProto(stg_ap_pp_fast) \
924 SymI_HasProto(stg_ap_ppv_fast) \
925 SymI_HasProto(stg_ap_ppp_fast) \
926 SymI_HasProto(stg_ap_pppv_fast) \
927 SymI_HasProto(stg_ap_pppp_fast) \
928 SymI_HasProto(stg_ap_ppppp_fast) \
929 SymI_HasProto(stg_ap_pppppp_fast) \
930 SymI_HasProto(stg_ap_1_upd_info) \
931 SymI_HasProto(stg_ap_2_upd_info) \
932 SymI_HasProto(stg_ap_3_upd_info) \
933 SymI_HasProto(stg_ap_4_upd_info) \
934 SymI_HasProto(stg_ap_5_upd_info) \
935 SymI_HasProto(stg_ap_6_upd_info) \
936 SymI_HasProto(stg_ap_7_upd_info) \
937 SymI_HasProto(stg_exit) \
938 SymI_HasProto(stg_sel_0_upd_info) \
939 SymI_HasProto(stg_sel_10_upd_info) \
940 SymI_HasProto(stg_sel_11_upd_info) \
941 SymI_HasProto(stg_sel_12_upd_info) \
942 SymI_HasProto(stg_sel_13_upd_info) \
943 SymI_HasProto(stg_sel_14_upd_info) \
944 SymI_HasProto(stg_sel_15_upd_info) \
945 SymI_HasProto(stg_sel_1_upd_info) \
946 SymI_HasProto(stg_sel_2_upd_info) \
947 SymI_HasProto(stg_sel_3_upd_info) \
948 SymI_HasProto(stg_sel_4_upd_info) \
949 SymI_HasProto(stg_sel_5_upd_info) \
950 SymI_HasProto(stg_sel_6_upd_info) \
951 SymI_HasProto(stg_sel_7_upd_info) \
952 SymI_HasProto(stg_sel_8_upd_info) \
953 SymI_HasProto(stg_sel_9_upd_info) \
954 SymI_HasProto(stg_upd_frame_info) \
955 SymI_HasProto(stg_bh_upd_frame_info) \
956 SymI_HasProto(suspendThread) \
957 SymI_HasProto(stg_takeMVarzh) \
958 SymI_HasProto(stg_threadStatuszh) \
959 SymI_HasProto(stg_tryPutMVarzh) \
960 SymI_HasProto(stg_tryTakeMVarzh) \
961 SymI_HasProto(stg_unmaskAsyncExceptionszh) \
962 SymI_HasProto(unloadObj) \
963 SymI_HasProto(stg_unsafeThawArrayzh) \
964 SymI_HasProto(stg_waitReadzh) \
965 SymI_HasProto(stg_waitWritezh) \
966 SymI_HasProto(stg_writeTVarzh) \
967 SymI_HasProto(stg_yieldzh) \
968 SymI_NeedsProto(stg_interp_constr_entry) \
969 SymI_HasProto(stg_arg_bitmaps) \
970 SymI_HasProto(alloc_blocks_lim) \
972 SymI_HasProto(allocate) \
973 SymI_HasProto(allocateExec) \
974 SymI_HasProto(freeExec) \
975 SymI_HasProto(getAllocations) \
976 SymI_HasProto(revertCAFs) \
977 SymI_HasProto(RtsFlags) \
978 SymI_NeedsProto(rts_breakpoint_io_action) \
979 SymI_NeedsProto(rts_stop_next_breakpoint) \
980 SymI_NeedsProto(rts_stop_on_exception) \
981 SymI_HasProto(stopTimer) \
982 SymI_HasProto(n_capabilities) \
983 SymI_HasProto(stg_traceCcszh) \
984 SymI_HasProto(stg_traceEventzh) \
985 RTS_USER_SIGNALS_SYMBOLS \
989 // 64-bit support functions in libgcc.a
990 #if defined(__GNUC__) && SIZEOF_VOID_P <= 4
991 #define RTS_LIBGCC_SYMBOLS \
992 SymI_NeedsProto(__divdi3) \
993 SymI_NeedsProto(__udivdi3) \
994 SymI_NeedsProto(__moddi3) \
995 SymI_NeedsProto(__umoddi3) \
996 SymI_NeedsProto(__muldi3) \
997 SymI_NeedsProto(__ashldi3) \
998 SymI_NeedsProto(__ashrdi3) \
999 SymI_NeedsProto(__lshrdi3)
1001 #define RTS_LIBGCC_SYMBOLS
1004 #if defined(darwin_HOST_OS) && defined(powerpc_HOST_ARCH)
1005 // Symbols that don't have a leading underscore
1006 // on Mac OS X. They have to receive special treatment,
1007 // see machoInitSymbolsWithoutUnderscore()
1008 #define RTS_MACHO_NOUNDERLINE_SYMBOLS \
1009 SymI_NeedsProto(saveFP) \
1010 SymI_NeedsProto(restFP)
1013 /* entirely bogus claims about types of these symbols */
1014 #define SymI_NeedsProto(vvv) extern void vvv(void);
1015 #if defined(__PIC__) && defined(mingw32_HOST_OS)
1016 #define SymE_HasProto(vvv) SymE_HasProto(vvv);
1017 #define SymE_NeedsProto(vvv) extern void _imp__ ## vvv (void);
1019 #define SymE_NeedsProto(vvv) SymI_NeedsProto(vvv);
1020 #define SymE_HasProto(vvv) SymI_HasProto(vvv)
1022 #define SymI_HasProto(vvv) /**/
1023 #define SymI_HasProto_redirect(vvv,xxx) /**/
1026 RTS_POSIX_ONLY_SYMBOLS
1027 RTS_MINGW_ONLY_SYMBOLS
1028 RTS_CYGWIN_ONLY_SYMBOLS
1029 RTS_DARWIN_ONLY_SYMBOLS
1032 #undef SymI_NeedsProto
1033 #undef SymI_HasProto
1034 #undef SymI_HasProto_redirect
1035 #undef SymE_HasProto
1036 #undef SymE_NeedsProto
1038 #ifdef LEADING_UNDERSCORE
1039 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
1041 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
1044 #define SymI_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1046 #define SymE_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1047 (void*)DLL_IMPORT_DATA_REF(vvv) },
1049 #define SymI_NeedsProto(vvv) SymI_HasProto(vvv)
1050 #define SymE_NeedsProto(vvv) SymE_HasProto(vvv)
1052 // SymI_HasProto_redirect allows us to redirect references to one symbol to
1053 // another symbol. See newCAF/newDynCAF for an example.
1054 #define SymI_HasProto_redirect(vvv,xxx) \
1055 { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1058 static RtsSymbolVal rtsSyms[] = {
1061 RTS_POSIX_ONLY_SYMBOLS
1062 RTS_MINGW_ONLY_SYMBOLS
1063 RTS_CYGWIN_ONLY_SYMBOLS
1064 RTS_DARWIN_ONLY_SYMBOLS
1067 #if defined(darwin_HOST_OS) && defined(i386_HOST_ARCH)
1068 // dyld stub code contains references to this,
1069 // but it should never be called because we treat
1070 // lazy pointers as nonlazy.
1071 { "dyld_stub_binding_helper", (void*)0xDEADBEEF },
1073 { 0, 0 } /* sentinel */
1078 /* -----------------------------------------------------------------------------
1079 * Insert symbols into hash tables, checking for duplicates.
1082 static void ghciInsertStrHashTable ( char* obj_name,
1088 if (lookupHashTable(table, (StgWord)key) == NULL)
1090 insertStrHashTable(table, (StgWord)key, data);
1095 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
1097 "whilst processing object file\n"
1099 "This could be caused by:\n"
1100 " * Loading two different object files which export the same symbol\n"
1101 " * Specifying the same object file twice on the GHCi command line\n"
1102 " * An incorrect `package.conf' entry, causing some object to be\n"
1104 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
1111 /* -----------------------------------------------------------------------------
1112 * initialize the object linker
1116 static int linker_init_done = 0 ;
1118 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1119 static void *dl_prog_handle;
1120 static regex_t re_invalid;
1121 static regex_t re_realso;
1123 static Mutex dl_mutex; // mutex to protect dlopen/dlerror critical section
1131 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1135 /* Make initLinker idempotent, so we can call it
1136 before evey relevant operation; that means we
1137 don't need to initialise the linker separately */
1138 if (linker_init_done == 1) { return; } else {
1139 linker_init_done = 1;
1142 #if defined(THREADED_RTS) && (defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO))
1143 initMutex(&dl_mutex);
1145 stablehash = allocStrHashTable();
1146 symhash = allocStrHashTable();
1148 /* populate the symbol table with stuff from the RTS */
1149 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
1150 ghciInsertStrHashTable("(GHCi built-in symbols)",
1151 symhash, sym->lbl, sym->addr);
1153 # if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH)
1154 machoInitSymbolsWithoutUnderscore();
1157 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1158 # if defined(RTLD_DEFAULT)
1159 dl_prog_handle = RTLD_DEFAULT;
1161 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
1162 # endif /* RTLD_DEFAULT */
1164 compileResult = regcomp(&re_invalid,
1165 "(([^ \t()])+\\.so([^ \t:()])*):([ \t])*invalid ELF header",
1167 ASSERT( compileResult == 0 );
1168 compileResult = regcomp(&re_realso,
1169 "GROUP *\\( *(([^ )])+)",
1171 ASSERT( compileResult == 0 );
1174 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
1175 if (RtsFlags.MiscFlags.linkerMemBase != 0) {
1176 // User-override for mmap_32bit_base
1177 mmap_32bit_base = (void*)RtsFlags.MiscFlags.linkerMemBase;
1181 #if defined(mingw32_HOST_OS)
1183 * These two libraries cause problems when added to the static link,
1184 * but are necessary for resolving symbols in GHCi, hence we load
1185 * them manually here.
1193 exitLinker( void ) {
1194 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1195 if (linker_init_done == 1) {
1196 regfree(&re_invalid);
1197 regfree(&re_realso);
1199 closeMutex(&dl_mutex);
1205 /* -----------------------------------------------------------------------------
1206 * Loading DLL or .so dynamic libraries
1207 * -----------------------------------------------------------------------------
1209 * Add a DLL from which symbols may be found. In the ELF case, just
1210 * do RTLD_GLOBAL-style add, so no further messing around needs to
1211 * happen in order that symbols in the loaded .so are findable --
1212 * lookupSymbol() will subsequently see them by dlsym on the program's
1213 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
1215 * In the PEi386 case, open the DLLs and put handles to them in a
1216 * linked list. When looking for a symbol, try all handles in the
1217 * list. This means that we need to load even DLLs that are guaranteed
1218 * to be in the ghc.exe image already, just so we can get a handle
1219 * to give to loadSymbol, so that we can find the symbols. For such
1220 * libraries, the LoadLibrary call should be a no-op except for returning
1225 #if defined(OBJFORMAT_PEi386)
1226 /* A record for storing handles into DLLs. */
1231 struct _OpenedDLL* next;
1236 /* A list thereof. */
1237 static OpenedDLL* opened_dlls = NULL;
1240 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1243 internal_dlopen(const char *dll_name)
1249 // omitted: RTLD_NOW
1250 // see http://www.haskell.org/pipermail/cvs-ghc/2007-September/038570.html
1252 debugBelch("internal_dlopen: dll_name = '%s'\n", dll_name));
1254 //-------------- Begin critical section ------------------
1255 // This critical section is necessary because dlerror() is not
1256 // required to be reentrant (see POSIX -- IEEE Std 1003.1-2008)
1257 // Also, the error message returned must be copied to preserve it
1260 ACQUIRE_LOCK(&dl_mutex);
1261 hdl = dlopen(dll_name, RTLD_LAZY | RTLD_GLOBAL);
1265 /* dlopen failed; return a ptr to the error msg. */
1267 if (errmsg == NULL) errmsg = "addDLL: unknown error";
1268 errmsg_copy = stgMallocBytes(strlen(errmsg)+1, "addDLL");
1269 strcpy(errmsg_copy, errmsg);
1270 errmsg = errmsg_copy;
1272 RELEASE_LOCK(&dl_mutex);
1273 //--------------- End critical section -------------------
1280 addDLL( char *dll_name )
1282 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1283 /* ------------------- ELF DLL loader ------------------- */
1286 regmatch_t match[NMATCH];
1289 size_t match_length;
1290 #define MAXLINE 1000
1296 IF_DEBUG(linker, debugBelch("addDLL: dll_name = '%s'\n", dll_name));
1297 errmsg = internal_dlopen(dll_name);
1299 if (errmsg == NULL) {
1303 // GHC Trac ticket #2615
1304 // On some systems (e.g., Gentoo Linux) dynamic files (e.g. libc.so)
1305 // contain linker scripts rather than ELF-format object code. This
1306 // code handles the situation by recognizing the real object code
1307 // file name given in the linker script.
1309 // If an "invalid ELF header" error occurs, it is assumed that the
1310 // .so file contains a linker script instead of ELF object code.
1311 // In this case, the code looks for the GROUP ( ... ) linker
1312 // directive. If one is found, the first file name inside the
1313 // parentheses is treated as the name of a dynamic library and the
1314 // code attempts to dlopen that file. If this is also unsuccessful,
1315 // an error message is returned.
1317 // see if the error message is due to an invalid ELF header
1318 IF_DEBUG(linker, debugBelch("errmsg = '%s'\n", errmsg));
1319 result = regexec(&re_invalid, errmsg, (size_t) NMATCH, match, 0);
1320 IF_DEBUG(linker, debugBelch("result = %i\n", result));
1322 // success -- try to read the named file as a linker script
1323 match_length = (size_t) stg_min((match[1].rm_eo - match[1].rm_so),
1325 strncpy(line, (errmsg+(match[1].rm_so)),match_length);
1326 line[match_length] = '\0'; // make sure string is null-terminated
1327 IF_DEBUG(linker, debugBelch ("file name = '%s'\n", line));
1328 if ((fp = fopen(line, "r")) == NULL) {
1329 return errmsg; // return original error if open fails
1331 // try to find a GROUP ( ... ) command
1332 while (fgets(line, MAXLINE, fp) != NULL) {
1333 IF_DEBUG(linker, debugBelch("input line = %s", line));
1334 if (regexec(&re_realso, line, (size_t) NMATCH, match, 0) == 0) {
1335 // success -- try to dlopen the first named file
1336 IF_DEBUG(linker, debugBelch("match%s\n",""));
1337 line[match[1].rm_eo] = '\0';
1338 errmsg = internal_dlopen(line+match[1].rm_so);
1341 // if control reaches here, no GROUP ( ... ) directive was found
1342 // and the original error message is returned to the caller
1348 # elif defined(OBJFORMAT_PEi386)
1349 /* ------------------- Win32 DLL loader ------------------- */
1357 /* debugBelch("\naddDLL; dll_name = `%s'\n", dll_name); */
1359 /* See if we've already got it, and ignore if so. */
1360 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
1361 if (0 == strcmp(o_dll->name, dll_name))
1365 /* The file name has no suffix (yet) so that we can try
1366 both foo.dll and foo.drv
1368 The documentation for LoadLibrary says:
1369 If no file name extension is specified in the lpFileName
1370 parameter, the default library extension .dll is
1371 appended. However, the file name string can include a trailing
1372 point character (.) to indicate that the module name has no
1375 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
1376 sprintf(buf, "%s.DLL", dll_name);
1377 instance = LoadLibrary(buf);
1378 if (instance == NULL) {
1379 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1380 // KAA: allow loading of drivers (like winspool.drv)
1381 sprintf(buf, "%s.DRV", dll_name);
1382 instance = LoadLibrary(buf);
1383 if (instance == NULL) {
1384 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1385 // #1883: allow loading of unix-style libfoo.dll DLLs
1386 sprintf(buf, "lib%s.DLL", dll_name);
1387 instance = LoadLibrary(buf);
1388 if (instance == NULL) {
1395 /* Add this DLL to the list of DLLs in which to search for symbols. */
1396 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
1397 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
1398 strcpy(o_dll->name, dll_name);
1399 o_dll->instance = instance;
1400 o_dll->next = opened_dlls;
1401 opened_dlls = o_dll;
1407 sysErrorBelch(dll_name);
1409 /* LoadLibrary failed; return a ptr to the error msg. */
1410 return "addDLL: could not load DLL";
1413 barf("addDLL: not implemented on this platform");
1417 /* -----------------------------------------------------------------------------
1418 * insert a stable symbol in the hash table
1422 insertStableSymbol(char* obj_name, char* key, StgPtr p)
1424 ghciInsertStrHashTable(obj_name, stablehash, key, getStablePtr(p));
1428 /* -----------------------------------------------------------------------------
1429 * insert a symbol in the hash table
1432 insertSymbol(char* obj_name, char* key, void* data)
1434 ghciInsertStrHashTable(obj_name, symhash, key, data);
1437 /* -----------------------------------------------------------------------------
1438 * lookup a symbol in the hash table
1441 lookupSymbol( char *lbl )
1445 ASSERT(symhash != NULL);
1446 val = lookupStrHashTable(symhash, lbl);
1449 # if defined(OBJFORMAT_ELF)
1450 return dlsym(dl_prog_handle, lbl);
1451 # elif defined(OBJFORMAT_MACHO)
1453 /* On OS X 10.3 and later, we use dlsym instead of the old legacy
1456 HACK: On OS X, global symbols are prefixed with an underscore.
1457 However, dlsym wants us to omit the leading underscore from the
1458 symbol name. For now, we simply strip it off here (and ONLY
1461 ASSERT(lbl[0] == '_');
1462 return dlsym(dl_prog_handle, lbl+1);
1464 if(NSIsSymbolNameDefined(lbl)) {
1465 NSSymbol symbol = NSLookupAndBindSymbol(lbl);
1466 return NSAddressOfSymbol(symbol);
1470 # endif /* HAVE_DLFCN_H */
1471 # elif defined(OBJFORMAT_PEi386)
1474 sym = lookupSymbolInDLLs((unsigned char*)lbl);
1475 if (sym != NULL) { return sym; };
1477 // Also try looking up the symbol without the @N suffix. Some
1478 // DLLs have the suffixes on their symbols, some don't.
1479 zapTrailingAtSign ( (unsigned char*)lbl );
1480 sym = lookupSymbolInDLLs((unsigned char*)lbl);
1481 if (sym != NULL) { return sym; };
1493 /* -----------------------------------------------------------------------------
1494 * Debugging aid: look in GHCi's object symbol tables for symbols
1495 * within DELTA bytes of the specified address, and show their names.
1498 void ghci_enquire ( char* addr );
1500 void ghci_enquire ( char* addr )
1505 const int DELTA = 64;
1510 for (oc = objects; oc; oc = oc->next) {
1511 for (i = 0; i < oc->n_symbols; i++) {
1512 sym = oc->symbols[i];
1513 if (sym == NULL) continue;
1516 a = lookupStrHashTable(symhash, sym);
1519 // debugBelch("ghci_enquire: can't find %s\n", sym);
1521 else if (addr-DELTA <= a && a <= addr+DELTA) {
1522 debugBelch("%p + %3d == `%s'\n", addr, (int)(a - addr), sym);
1530 #define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))
1533 mmapForLinker (size_t bytes, nat flags, int fd)
1535 void *map_addr = NULL;
1538 static nat fixed = 0;
1540 pagesize = getpagesize();
1541 size = ROUND_UP(bytes, pagesize);
1543 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
1546 if (mmap_32bit_base != 0) {
1547 map_addr = mmap_32bit_base;
1551 result = mmap(map_addr, size, PROT_EXEC|PROT_READ|PROT_WRITE,
1552 MAP_PRIVATE|TRY_MAP_32BIT|fixed|flags, fd, 0);
1554 if (result == MAP_FAILED) {
1555 sysErrorBelch("mmap %lu bytes at %p",(lnat)size,map_addr);
1556 errorBelch("Try specifying an address with +RTS -xm<addr> -RTS");
1557 stg_exit(EXIT_FAILURE);
1560 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
1561 if (mmap_32bit_base != 0) {
1562 if (result == map_addr) {
1563 mmap_32bit_base = (StgWord8*)map_addr + size;
1565 if ((W_)result > 0x80000000) {
1566 // oops, we were given memory over 2Gb
1567 #if defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS)
1568 // Some platforms require MAP_FIXED. This is normally
1569 // a bad idea, because MAP_FIXED will overwrite
1570 // existing mappings.
1571 munmap(result,size);
1575 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);
1578 // hmm, we were given memory somewhere else, but it's
1579 // still under 2Gb so we can use it. Next time, ask
1580 // for memory right after the place we just got some
1581 mmap_32bit_base = (StgWord8*)result + size;
1585 if ((W_)result > 0x80000000) {
1586 // oops, we were given memory over 2Gb
1587 // ... try allocating memory somewhere else?;
1588 debugTrace(DEBUG_linker,"MAP_32BIT didn't work; gave us %lu bytes at 0x%p", bytes, result);
1589 munmap(result, size);
1591 // Set a base address and try again... (guess: 1Gb)
1592 mmap_32bit_base = (void*)0x40000000;
1602 /* -----------------------------------------------------------------------------
1603 * Load an obj (populate the global symbol table, but don't resolve yet)
1605 * Returns: 1 if ok, 0 on error.
1608 loadObj( char *path )
1618 IF_DEBUG(linker, debugBelch("loadObj %s\n", path));
1621 /* debugBelch("loadObj %s\n", path ); */
1623 /* Check that we haven't already loaded this object.
1624 Ignore requests to load multiple times */
1628 for (o = objects; o; o = o->next) {
1629 if (0 == strcmp(o->fileName, path)) {
1631 break; /* don't need to search further */
1635 IF_DEBUG(linker, debugBelch(
1636 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1637 "same object file twice:\n"
1639 "GHCi will ignore this, but be warned.\n"
1641 return 1; /* success */
1645 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
1647 # if defined(OBJFORMAT_ELF)
1648 oc->formatName = "ELF";
1649 # elif defined(OBJFORMAT_PEi386)
1650 oc->formatName = "PEi386";
1651 # elif defined(OBJFORMAT_MACHO)
1652 oc->formatName = "Mach-O";
1655 barf("loadObj: not implemented on this platform");
1658 r = stat(path, &st);
1660 IF_DEBUG(linker, debugBelch("File doesn't exist\n"));
1664 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1665 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1666 strcpy(oc->fileName, path);
1668 oc->fileSize = st.st_size;
1670 oc->sections = NULL;
1671 oc->proddables = NULL;
1673 /* chain it onto the list of objects */
1678 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1680 #if defined(openbsd_HOST_OS)
1681 fd = open(path, O_RDONLY, S_IRUSR);
1683 fd = open(path, O_RDONLY);
1686 barf("loadObj: can't open `%s'", path);
1688 oc->image = mmapForLinker(oc->fileSize, 0, fd);
1692 #else /* !USE_MMAP */
1693 /* load the image into memory */
1694 f = fopen(path, "rb");
1696 barf("loadObj: can't read `%s'", path);
1698 # if defined(mingw32_HOST_OS)
1699 // TODO: We would like to use allocateExec here, but allocateExec
1700 // cannot currently allocate blocks large enough.
1701 oc->image = VirtualAlloc(NULL, oc->fileSize, MEM_RESERVE | MEM_COMMIT,
1702 PAGE_EXECUTE_READWRITE);
1703 # elif defined(darwin_HOST_OS)
1704 // In a Mach-O .o file, all sections can and will be misaligned
1705 // if the total size of the headers is not a multiple of the
1706 // desired alignment. This is fine for .o files that only serve
1707 // as input for the static linker, but it's not fine for us,
1708 // as SSE (used by gcc for floating point) and Altivec require
1709 // 16-byte alignment.
1710 // We calculate the correct alignment from the header before
1711 // reading the file, and then we misalign oc->image on purpose so
1712 // that the actual sections end up aligned again.
1713 oc->misalignment = machoGetMisalignment(f);
1714 oc->image = stgMallocBytes(oc->fileSize + oc->misalignment, "loadObj(image)");
1715 oc->image += oc->misalignment;
1717 oc->image = stgMallocBytes(oc->fileSize, "loadObj(image)");
1722 n = fread ( oc->image, 1, oc->fileSize, f );
1723 if (n != oc->fileSize)
1724 barf("loadObj: error whilst reading `%s'", path);
1727 #endif /* USE_MMAP */
1729 # if defined(OBJFORMAT_MACHO) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1730 r = ocAllocateSymbolExtras_MachO ( oc );
1732 IF_DEBUG(linker, debugBelch("ocAllocateSymbolExtras_MachO failed\n"));
1735 # elif defined(OBJFORMAT_ELF) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1736 r = ocAllocateSymbolExtras_ELF ( oc );
1738 IF_DEBUG(linker, debugBelch("ocAllocateSymbolExtras_ELF failed\n"));
1743 /* verify the in-memory image */
1744 # if defined(OBJFORMAT_ELF)
1745 r = ocVerifyImage_ELF ( oc );
1746 # elif defined(OBJFORMAT_PEi386)
1747 r = ocVerifyImage_PEi386 ( oc );
1748 # elif defined(OBJFORMAT_MACHO)
1749 r = ocVerifyImage_MachO ( oc );
1751 barf("loadObj: no verify method");
1754 IF_DEBUG(linker, debugBelch("ocVerifyImage_* failed\n"));
1758 /* build the symbol list for this image */
1759 # if defined(OBJFORMAT_ELF)
1760 r = ocGetNames_ELF ( oc );
1761 # elif defined(OBJFORMAT_PEi386)
1762 r = ocGetNames_PEi386 ( oc );
1763 # elif defined(OBJFORMAT_MACHO)
1764 r = ocGetNames_MachO ( oc );
1766 barf("loadObj: no getNames method");
1769 IF_DEBUG(linker, debugBelch("ocGetNames_* failed\n"));
1773 /* loaded, but not resolved yet */
1774 oc->status = OBJECT_LOADED;
1779 /* -----------------------------------------------------------------------------
1780 * resolve all the currently unlinked objects in memory
1782 * Returns: 1 if ok, 0 on error.
1792 for (oc = objects; oc; oc = oc->next) {
1793 if (oc->status != OBJECT_RESOLVED) {
1794 # if defined(OBJFORMAT_ELF)
1795 r = ocResolve_ELF ( oc );
1796 # elif defined(OBJFORMAT_PEi386)
1797 r = ocResolve_PEi386 ( oc );
1798 # elif defined(OBJFORMAT_MACHO)
1799 r = ocResolve_MachO ( oc );
1801 barf("resolveObjs: not implemented on this platform");
1803 if (!r) { return r; }
1804 oc->status = OBJECT_RESOLVED;
1810 /* -----------------------------------------------------------------------------
1811 * delete an object from the pool
1814 unloadObj( char *path )
1816 ObjectCode *oc, *prev;
1818 ASSERT(symhash != NULL);
1819 ASSERT(objects != NULL);
1824 for (oc = objects; oc; prev = oc, oc = oc->next) {
1825 if (!strcmp(oc->fileName,path)) {
1827 /* Remove all the mappings for the symbols within this
1832 for (i = 0; i < oc->n_symbols; i++) {
1833 if (oc->symbols[i] != NULL) {
1834 removeStrHashTable(symhash, oc->symbols[i], NULL);
1842 prev->next = oc->next;
1845 // We're going to leave this in place, in case there are
1846 // any pointers from the heap into it:
1847 // #ifdef mingw32_HOST_OS
1848 // VirtualFree(oc->image);
1850 // stgFree(oc->image);
1852 stgFree(oc->fileName);
1853 stgFree(oc->symbols);
1854 stgFree(oc->sections);
1860 errorBelch("unloadObj: can't find `%s' to unload", path);
1864 /* -----------------------------------------------------------------------------
1865 * Sanity checking. For each ObjectCode, maintain a list of address ranges
1866 * which may be prodded during relocation, and abort if we try and write
1867 * outside any of these.
1869 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
1872 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
1873 /* debugBelch("aPB %p %p %d\n", oc, start, size); */
1877 pb->next = oc->proddables;
1878 oc->proddables = pb;
1881 static void checkProddableBlock ( ObjectCode* oc, void* addr )
1884 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
1885 char* s = (char*)(pb->start);
1886 char* e = s + pb->size - 1;
1887 char* a = (char*)addr;
1888 /* Assumes that the biggest fixup involves a 4-byte write. This
1889 probably needs to be changed to 8 (ie, +7) on 64-bit
1891 if (a >= s && (a+3) <= e) return;
1893 barf("checkProddableBlock: invalid fixup in runtime linker");
1896 /* -----------------------------------------------------------------------------
1897 * Section management.
1899 static void addSection ( ObjectCode* oc, SectionKind kind,
1900 void* start, void* end )
1902 Section* s = stgMallocBytes(sizeof(Section), "addSection");
1906 s->next = oc->sections;
1909 debugBelch("addSection: %p-%p (size %d), kind %d\n",
1910 start, ((char*)end)-1, end - start + 1, kind );
1915 /* --------------------------------------------------------------------------
1917 * This is about allocating a small chunk of memory for every symbol in the
1918 * object file. We make sure that the SymboLExtras are always "in range" of
1919 * limited-range PC-relative instructions on various platforms by allocating
1920 * them right next to the object code itself.
1923 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
1926 ocAllocateSymbolExtras
1928 Allocate additional space at the end of the object file image to make room
1929 for jump islands (powerpc, x86_64) and GOT entries (x86_64).
1931 PowerPC relative branch instructions have a 24 bit displacement field.
1932 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
1933 If a particular imported symbol is outside this range, we have to redirect
1934 the jump to a short piece of new code that just loads the 32bit absolute
1935 address and jumps there.
1936 On x86_64, PC-relative jumps and PC-relative accesses to the GOT are limited
1939 This function just allocates space for one SymbolExtra for every
1940 undefined symbol in the object file. The code for the jump islands is
1941 filled in by makeSymbolExtra below.
1944 static int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first )
1951 int misalignment = 0;
1952 #ifdef darwin_HOST_OS
1953 misalignment = oc->misalignment;
1959 // round up to the nearest 4
1960 aligned = (oc->fileSize + 3) & ~3;
1963 pagesize = getpagesize();
1964 n = ROUND_UP( oc->fileSize, pagesize );
1965 m = ROUND_UP( aligned + sizeof (SymbolExtra) * count, pagesize );
1967 /* we try to use spare space at the end of the last page of the
1968 * image for the jump islands, but if there isn't enough space
1969 * then we have to map some (anonymously, remembering MAP_32BIT).
1971 if( m > n ) // we need to allocate more pages
1973 oc->symbol_extras = mmapForLinker(sizeof(SymbolExtra) * count,
1978 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1981 oc->image -= misalignment;
1982 oc->image = stgReallocBytes( oc->image,
1984 aligned + sizeof (SymbolExtra) * count,
1985 "ocAllocateSymbolExtras" );
1986 oc->image += misalignment;
1988 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1989 #endif /* USE_MMAP */
1991 memset( oc->symbol_extras, 0, sizeof (SymbolExtra) * count );
1994 oc->symbol_extras = NULL;
1996 oc->first_symbol_extra = first;
1997 oc->n_symbol_extras = count;
2002 static SymbolExtra* makeSymbolExtra( ObjectCode* oc,
2003 unsigned long symbolNumber,
2004 unsigned long target )
2008 ASSERT( symbolNumber >= oc->first_symbol_extra
2009 && symbolNumber - oc->first_symbol_extra < oc->n_symbol_extras);
2011 extra = &oc->symbol_extras[symbolNumber - oc->first_symbol_extra];
2013 #ifdef powerpc_HOST_ARCH
2014 // lis r12, hi16(target)
2015 extra->jumpIsland.lis_r12 = 0x3d80;
2016 extra->jumpIsland.hi_addr = target >> 16;
2018 // ori r12, r12, lo16(target)
2019 extra->jumpIsland.ori_r12_r12 = 0x618c;
2020 extra->jumpIsland.lo_addr = target & 0xffff;
2023 extra->jumpIsland.mtctr_r12 = 0x7d8903a6;
2026 extra->jumpIsland.bctr = 0x4e800420;
2028 #ifdef x86_64_HOST_ARCH
2030 static uint8_t jmp[] = { 0xFF, 0x25, 0xF2, 0xFF, 0xFF, 0xFF };
2031 extra->addr = target;
2032 memcpy(extra->jumpIsland, jmp, 6);
2040 /* --------------------------------------------------------------------------
2041 * PowerPC specifics (instruction cache flushing)
2042 * ------------------------------------------------------------------------*/
2044 #ifdef powerpc_HOST_ARCH
2046 ocFlushInstructionCache
2048 Flush the data & instruction caches.
2049 Because the PPC has split data/instruction caches, we have to
2050 do that whenever we modify code at runtime.
2053 static void ocFlushInstructionCache( ObjectCode *oc )
2055 int n = (oc->fileSize + sizeof( SymbolExtra ) * oc->n_symbol_extras + 3) / 4;
2056 unsigned long *p = (unsigned long *) oc->image;
2060 __asm__ volatile ( "dcbf 0,%0\n\t"
2068 __asm__ volatile ( "sync\n\t"
2074 /* --------------------------------------------------------------------------
2075 * PEi386 specifics (Win32 targets)
2076 * ------------------------------------------------------------------------*/
2078 /* The information for this linker comes from
2079 Microsoft Portable Executable
2080 and Common Object File Format Specification
2081 revision 5.1 January 1998
2082 which SimonM says comes from the MS Developer Network CDs.
2084 It can be found there (on older CDs), but can also be found
2087 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
2089 (this is Rev 6.0 from February 1999).
2091 Things move, so if that fails, try searching for it via
2093 http://www.google.com/search?q=PE+COFF+specification
2095 The ultimate reference for the PE format is the Winnt.h
2096 header file that comes with the Platform SDKs; as always,
2097 implementations will drift wrt their documentation.
2099 A good background article on the PE format is Matt Pietrek's
2100 March 1994 article in Microsoft System Journal (MSJ)
2101 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
2102 Win32 Portable Executable File Format." The info in there
2103 has recently been updated in a two part article in
2104 MSDN magazine, issues Feb and March 2002,
2105 "Inside Windows: An In-Depth Look into the Win32 Portable
2106 Executable File Format"
2108 John Levine's book "Linkers and Loaders" contains useful
2113 #if defined(OBJFORMAT_PEi386)
2117 typedef unsigned char UChar;
2118 typedef unsigned short UInt16;
2119 typedef unsigned int UInt32;
2126 UInt16 NumberOfSections;
2127 UInt32 TimeDateStamp;
2128 UInt32 PointerToSymbolTable;
2129 UInt32 NumberOfSymbols;
2130 UInt16 SizeOfOptionalHeader;
2131 UInt16 Characteristics;
2135 #define sizeof_COFF_header 20
2142 UInt32 VirtualAddress;
2143 UInt32 SizeOfRawData;
2144 UInt32 PointerToRawData;
2145 UInt32 PointerToRelocations;
2146 UInt32 PointerToLinenumbers;
2147 UInt16 NumberOfRelocations;
2148 UInt16 NumberOfLineNumbers;
2149 UInt32 Characteristics;
2153 #define sizeof_COFF_section 40
2160 UInt16 SectionNumber;
2163 UChar NumberOfAuxSymbols;
2167 #define sizeof_COFF_symbol 18
2172 UInt32 VirtualAddress;
2173 UInt32 SymbolTableIndex;
2178 #define sizeof_COFF_reloc 10
2181 /* From PE spec doc, section 3.3.2 */
2182 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
2183 windows.h -- for the same purpose, but I want to know what I'm
2185 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
2186 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
2187 #define MYIMAGE_FILE_DLL 0x2000
2188 #define MYIMAGE_FILE_SYSTEM 0x1000
2189 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
2190 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
2191 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
2193 /* From PE spec doc, section 5.4.2 and 5.4.4 */
2194 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
2195 #define MYIMAGE_SYM_CLASS_STATIC 3
2196 #define MYIMAGE_SYM_UNDEFINED 0
2198 /* From PE spec doc, section 4.1 */
2199 #define MYIMAGE_SCN_CNT_CODE 0x00000020
2200 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
2201 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
2203 /* From PE spec doc, section 5.2.1 */
2204 #define MYIMAGE_REL_I386_DIR32 0x0006
2205 #define MYIMAGE_REL_I386_REL32 0x0014
2208 /* We use myindex to calculate array addresses, rather than
2209 simply doing the normal subscript thing. That's because
2210 some of the above structs have sizes which are not
2211 a whole number of words. GCC rounds their sizes up to a
2212 whole number of words, which means that the address calcs
2213 arising from using normal C indexing or pointer arithmetic
2214 are just plain wrong. Sigh.
2217 myindex ( int scale, void* base, int index )
2220 ((UChar*)base) + scale * index;
2225 printName ( UChar* name, UChar* strtab )
2227 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2228 UInt32 strtab_offset = * (UInt32*)(name+4);
2229 debugBelch("%s", strtab + strtab_offset );
2232 for (i = 0; i < 8; i++) {
2233 if (name[i] == 0) break;
2234 debugBelch("%c", name[i] );
2241 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
2243 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2244 UInt32 strtab_offset = * (UInt32*)(name+4);
2245 strncpy ( (char*)dst, (char*)strtab+strtab_offset, dstSize );
2251 if (name[i] == 0) break;
2261 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
2264 /* If the string is longer than 8 bytes, look in the
2265 string table for it -- this will be correctly zero terminated.
2267 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2268 UInt32 strtab_offset = * (UInt32*)(name+4);
2269 return ((UChar*)strtab) + strtab_offset;
2271 /* Otherwise, if shorter than 8 bytes, return the original,
2272 which by defn is correctly terminated.
2274 if (name[7]==0) return name;
2275 /* The annoying case: 8 bytes. Copy into a temporary
2276 (which is never freed ...)
2278 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
2280 strncpy((char*)newstr,(char*)name,8);
2286 /* Just compares the short names (first 8 chars) */
2287 static COFF_section *
2288 findPEi386SectionCalled ( ObjectCode* oc, UChar* name )
2292 = (COFF_header*)(oc->image);
2293 COFF_section* sectab
2295 ((UChar*)(oc->image))
2296 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2298 for (i = 0; i < hdr->NumberOfSections; i++) {
2301 COFF_section* section_i
2303 myindex ( sizeof_COFF_section, sectab, i );
2304 n1 = (UChar*) &(section_i->Name);
2306 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
2307 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
2308 n1[6]==n2[6] && n1[7]==n2[7])
2317 zapTrailingAtSign ( UChar* sym )
2319 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
2321 if (sym[0] == 0) return;
2323 while (sym[i] != 0) i++;
2326 while (j > 0 && my_isdigit(sym[j])) j--;
2327 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
2332 lookupSymbolInDLLs ( UChar *lbl )
2337 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
2338 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
2340 if (lbl[0] == '_') {
2341 /* HACK: if the name has an initial underscore, try stripping
2342 it off & look that up first. I've yet to verify whether there's
2343 a Rule that governs whether an initial '_' *should always* be
2344 stripped off when mapping from import lib name to the DLL name.
2346 sym = GetProcAddress(o_dll->instance, (char*)(lbl+1));
2348 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
2352 sym = GetProcAddress(o_dll->instance, (char*)lbl);
2354 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
2363 ocVerifyImage_PEi386 ( ObjectCode* oc )
2368 COFF_section* sectab;
2369 COFF_symbol* symtab;
2371 /* debugBelch("\nLOADING %s\n", oc->fileName); */
2372 hdr = (COFF_header*)(oc->image);
2373 sectab = (COFF_section*) (
2374 ((UChar*)(oc->image))
2375 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2377 symtab = (COFF_symbol*) (
2378 ((UChar*)(oc->image))
2379 + hdr->PointerToSymbolTable
2381 strtab = ((UChar*)symtab)
2382 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2384 if (hdr->Machine != 0x14c) {
2385 errorBelch("%s: Not x86 PEi386", oc->fileName);
2388 if (hdr->SizeOfOptionalHeader != 0) {
2389 errorBelch("%s: PEi386 with nonempty optional header", oc->fileName);
2392 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
2393 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
2394 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
2395 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
2396 errorBelch("%s: Not a PEi386 object file", oc->fileName);
2399 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
2400 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
2401 errorBelch("%s: Invalid PEi386 word size or endiannness: %d",
2403 (int)(hdr->Characteristics));
2406 /* If the string table size is way crazy, this might indicate that
2407 there are more than 64k relocations, despite claims to the
2408 contrary. Hence this test. */
2409 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
2411 if ( (*(UInt32*)strtab) > 600000 ) {
2412 /* Note that 600k has no special significance other than being
2413 big enough to handle the almost-2MB-sized lumps that
2414 constitute HSwin32*.o. */
2415 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
2420 /* No further verification after this point; only debug printing. */
2422 IF_DEBUG(linker, i=1);
2423 if (i == 0) return 1;
2425 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
2426 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
2427 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
2430 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
2431 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
2432 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
2433 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
2434 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
2435 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
2436 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
2438 /* Print the section table. */
2440 for (i = 0; i < hdr->NumberOfSections; i++) {
2442 COFF_section* sectab_i
2444 myindex ( sizeof_COFF_section, sectab, i );
2451 printName ( sectab_i->Name, strtab );
2461 sectab_i->VirtualSize,
2462 sectab_i->VirtualAddress,
2463 sectab_i->SizeOfRawData,
2464 sectab_i->PointerToRawData,
2465 sectab_i->NumberOfRelocations,
2466 sectab_i->PointerToRelocations,
2467 sectab_i->PointerToRawData
2469 reltab = (COFF_reloc*) (
2470 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2473 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2474 /* If the relocation field (a short) has overflowed, the
2475 * real count can be found in the first reloc entry.
2477 * See Section 4.1 (last para) of the PE spec (rev6.0).
2479 COFF_reloc* rel = (COFF_reloc*)
2480 myindex ( sizeof_COFF_reloc, reltab, 0 );
2481 noRelocs = rel->VirtualAddress;
2484 noRelocs = sectab_i->NumberOfRelocations;
2488 for (; j < noRelocs; j++) {
2490 COFF_reloc* rel = (COFF_reloc*)
2491 myindex ( sizeof_COFF_reloc, reltab, j );
2493 " type 0x%-4x vaddr 0x%-8x name `",
2495 rel->VirtualAddress );
2496 sym = (COFF_symbol*)
2497 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
2498 /* Hmm..mysterious looking offset - what's it for? SOF */
2499 printName ( sym->Name, strtab -10 );
2506 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
2507 debugBelch("---START of string table---\n");
2508 for (i = 4; i < *(Int32*)strtab; i++) {
2510 debugBelch("\n"); else
2511 debugBelch("%c", strtab[i] );
2513 debugBelch("--- END of string table---\n");
2518 COFF_symbol* symtab_i;
2519 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2520 symtab_i = (COFF_symbol*)
2521 myindex ( sizeof_COFF_symbol, symtab, i );
2527 printName ( symtab_i->Name, strtab );
2536 (Int32)(symtab_i->SectionNumber),
2537 (UInt32)symtab_i->Type,
2538 (UInt32)symtab_i->StorageClass,
2539 (UInt32)symtab_i->NumberOfAuxSymbols
2541 i += symtab_i->NumberOfAuxSymbols;
2551 ocGetNames_PEi386 ( ObjectCode* oc )
2554 COFF_section* sectab;
2555 COFF_symbol* symtab;
2562 hdr = (COFF_header*)(oc->image);
2563 sectab = (COFF_section*) (
2564 ((UChar*)(oc->image))
2565 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2567 symtab = (COFF_symbol*) (
2568 ((UChar*)(oc->image))
2569 + hdr->PointerToSymbolTable
2571 strtab = ((UChar*)(oc->image))
2572 + hdr->PointerToSymbolTable
2573 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2575 /* Allocate space for any (local, anonymous) .bss sections. */
2577 for (i = 0; i < hdr->NumberOfSections; i++) {
2580 COFF_section* sectab_i
2582 myindex ( sizeof_COFF_section, sectab, i );
2583 if (0 != strcmp((char*)sectab_i->Name, ".bss")) continue;
2584 /* sof 10/05: the PE spec text isn't too clear regarding what
2585 * the SizeOfRawData field is supposed to hold for object
2586 * file sections containing just uninitialized data -- for executables,
2587 * it is supposed to be zero; unclear what it's supposed to be
2588 * for object files. However, VirtualSize is guaranteed to be
2589 * zero for object files, which definitely suggests that SizeOfRawData
2590 * will be non-zero (where else would the size of this .bss section be
2591 * stored?) Looking at the COFF_section info for incoming object files,
2592 * this certainly appears to be the case.
2594 * => I suspect we've been incorrectly handling .bss sections in (relocatable)
2595 * object files up until now. This turned out to bite us with ghc-6.4.1's use
2596 * of gcc-3.4.x, which has started to emit initially-zeroed-out local 'static'
2597 * variable decls into to the .bss section. (The specific function in Q which
2598 * triggered this is libraries/base/cbits/dirUtils.c:__hscore_getFolderPath())
2600 if (sectab_i->VirtualSize == 0 && sectab_i->SizeOfRawData == 0) continue;
2601 /* This is a non-empty .bss section. Allocate zeroed space for
2602 it, and set its PointerToRawData field such that oc->image +
2603 PointerToRawData == addr_of_zeroed_space. */
2604 bss_sz = sectab_i->VirtualSize;
2605 if ( bss_sz < sectab_i->SizeOfRawData) { bss_sz = sectab_i->SizeOfRawData; }
2606 zspace = stgCallocBytes(1, bss_sz, "ocGetNames_PEi386(anonymous bss)");
2607 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
2608 addProddableBlock(oc, zspace, bss_sz);
2609 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
2612 /* Copy section information into the ObjectCode. */
2614 for (i = 0; i < hdr->NumberOfSections; i++) {
2620 = SECTIONKIND_OTHER;
2621 COFF_section* sectab_i
2623 myindex ( sizeof_COFF_section, sectab, i );
2624 IF_DEBUG(linker, debugBelch("section name = %s\n", sectab_i->Name ));
2627 /* I'm sure this is the Right Way to do it. However, the
2628 alternative of testing the sectab_i->Name field seems to
2629 work ok with Cygwin.
2631 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
2632 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
2633 kind = SECTIONKIND_CODE_OR_RODATA;
2636 if (0==strcmp(".text",(char*)sectab_i->Name) ||
2637 0==strcmp(".rdata",(char*)sectab_i->Name)||
2638 0==strcmp(".rodata",(char*)sectab_i->Name))
2639 kind = SECTIONKIND_CODE_OR_RODATA;
2640 if (0==strcmp(".data",(char*)sectab_i->Name) ||
2641 0==strcmp(".bss",(char*)sectab_i->Name))
2642 kind = SECTIONKIND_RWDATA;
2644 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
2645 sz = sectab_i->SizeOfRawData;
2646 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
2648 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
2649 end = start + sz - 1;
2651 if (kind == SECTIONKIND_OTHER
2652 /* Ignore sections called which contain stabs debugging
2654 && 0 != strcmp(".stab", (char*)sectab_i->Name)
2655 && 0 != strcmp(".stabstr", (char*)sectab_i->Name)
2656 /* ignore constructor section for now */
2657 && 0 != strcmp(".ctors", (char*)sectab_i->Name)
2658 /* ignore section generated from .ident */
2659 && 0!= strcmp("/4", (char*)sectab_i->Name)
2660 /* ignore unknown section that appeared in gcc 3.4.5(?) */
2661 && 0!= strcmp(".reloc", (char*)sectab_i->Name)
2663 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName);
2667 if (kind != SECTIONKIND_OTHER && end >= start) {
2668 addSection(oc, kind, start, end);
2669 addProddableBlock(oc, start, end - start + 1);
2673 /* Copy exported symbols into the ObjectCode. */
2675 oc->n_symbols = hdr->NumberOfSymbols;
2676 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2677 "ocGetNames_PEi386(oc->symbols)");
2678 /* Call me paranoid; I don't care. */
2679 for (i = 0; i < oc->n_symbols; i++)
2680 oc->symbols[i] = NULL;
2684 COFF_symbol* symtab_i;
2685 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2686 symtab_i = (COFF_symbol*)
2687 myindex ( sizeof_COFF_symbol, symtab, i );
2691 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2692 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2693 /* This symbol is global and defined, viz, exported */
2694 /* for MYIMAGE_SYMCLASS_EXTERNAL
2695 && !MYIMAGE_SYM_UNDEFINED,
2696 the address of the symbol is:
2697 address of relevant section + offset in section
2699 COFF_section* sectabent
2700 = (COFF_section*) myindex ( sizeof_COFF_section,
2702 symtab_i->SectionNumber-1 );
2703 addr = ((UChar*)(oc->image))
2704 + (sectabent->PointerToRawData
2708 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2709 && symtab_i->Value > 0) {
2710 /* This symbol isn't in any section at all, ie, global bss.
2711 Allocate zeroed space for it. */
2712 addr = stgCallocBytes(1, symtab_i->Value,
2713 "ocGetNames_PEi386(non-anonymous bss)");
2714 addSection(oc, SECTIONKIND_RWDATA, addr,
2715 ((UChar*)addr) + symtab_i->Value - 1);
2716 addProddableBlock(oc, addr, symtab_i->Value);
2717 /* debugBelch("BSS section at 0x%x\n", addr); */
2720 if (addr != NULL ) {
2721 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2722 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2723 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2724 ASSERT(i >= 0 && i < oc->n_symbols);
2725 /* cstring_from_COFF_symbol_name always succeeds. */
2726 oc->symbols[i] = (char*)sname;
2727 ghciInsertStrHashTable(oc->fileName, symhash, (char*)sname, addr);
2731 "IGNORING symbol %d\n"
2735 printName ( symtab_i->Name, strtab );
2744 (Int32)(symtab_i->SectionNumber),
2745 (UInt32)symtab_i->Type,
2746 (UInt32)symtab_i->StorageClass,
2747 (UInt32)symtab_i->NumberOfAuxSymbols
2752 i += symtab_i->NumberOfAuxSymbols;
2761 ocResolve_PEi386 ( ObjectCode* oc )
2764 COFF_section* sectab;
2765 COFF_symbol* symtab;
2775 /* ToDo: should be variable-sized? But is at least safe in the
2776 sense of buffer-overrun-proof. */
2778 /* debugBelch("resolving for %s\n", oc->fileName); */
2780 hdr = (COFF_header*)(oc->image);
2781 sectab = (COFF_section*) (
2782 ((UChar*)(oc->image))
2783 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2785 symtab = (COFF_symbol*) (
2786 ((UChar*)(oc->image))
2787 + hdr->PointerToSymbolTable
2789 strtab = ((UChar*)(oc->image))
2790 + hdr->PointerToSymbolTable
2791 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2793 for (i = 0; i < hdr->NumberOfSections; i++) {
2794 COFF_section* sectab_i
2796 myindex ( sizeof_COFF_section, sectab, i );
2799 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2802 /* Ignore sections called which contain stabs debugging
2804 if (0 == strcmp(".stab", (char*)sectab_i->Name)
2805 || 0 == strcmp(".stabstr", (char*)sectab_i->Name)
2806 || 0 == strcmp(".ctors", (char*)sectab_i->Name))
2809 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2810 /* If the relocation field (a short) has overflowed, the
2811 * real count can be found in the first reloc entry.
2813 * See Section 4.1 (last para) of the PE spec (rev6.0).
2815 * Nov2003 update: the GNU linker still doesn't correctly
2816 * handle the generation of relocatable object files with
2817 * overflown relocations. Hence the output to warn of potential
2820 COFF_reloc* rel = (COFF_reloc*)
2821 myindex ( sizeof_COFF_reloc, reltab, 0 );
2822 noRelocs = rel->VirtualAddress;
2824 /* 10/05: we now assume (and check for) a GNU ld that is capable
2825 * of handling object files with (>2^16) of relocs.
2828 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2833 noRelocs = sectab_i->NumberOfRelocations;
2838 for (; j < noRelocs; j++) {
2840 COFF_reloc* reltab_j
2842 myindex ( sizeof_COFF_reloc, reltab, j );
2844 /* the location to patch */
2846 ((UChar*)(oc->image))
2847 + (sectab_i->PointerToRawData
2848 + reltab_j->VirtualAddress
2849 - sectab_i->VirtualAddress )
2851 /* the existing contents of pP */
2853 /* the symbol to connect to */
2854 sym = (COFF_symbol*)
2855 myindex ( sizeof_COFF_symbol,
2856 symtab, reltab_j->SymbolTableIndex );
2859 "reloc sec %2d num %3d: type 0x%-4x "
2860 "vaddr 0x%-8x name `",
2862 (UInt32)reltab_j->Type,
2863 reltab_j->VirtualAddress );
2864 printName ( sym->Name, strtab );
2865 debugBelch("'\n" ));
2867 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2868 COFF_section* section_sym
2869 = findPEi386SectionCalled ( oc, sym->Name );
2871 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2874 S = ((UInt32)(oc->image))
2875 + (section_sym->PointerToRawData
2878 copyName ( sym->Name, strtab, symbol, 1000-1 );
2879 S = (UInt32) lookupSymbol( (char*)symbol );
2880 if ((void*)S != NULL) goto foundit;
2881 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
2885 checkProddableBlock(oc, pP);
2886 switch (reltab_j->Type) {
2887 case MYIMAGE_REL_I386_DIR32:
2890 case MYIMAGE_REL_I386_REL32:
2891 /* Tricky. We have to insert a displacement at
2892 pP which, when added to the PC for the _next_
2893 insn, gives the address of the target (S).
2894 Problem is to know the address of the next insn
2895 when we only know pP. We assume that this
2896 literal field is always the last in the insn,
2897 so that the address of the next insn is pP+4
2898 -- hence the constant 4.
2899 Also I don't know if A should be added, but so
2900 far it has always been zero.
2902 SOF 05/2005: 'A' (old contents of *pP) have been observed
2903 to contain values other than zero (the 'wx' object file
2904 that came with wxhaskell-0.9.4; dunno how it was compiled..).
2905 So, add displacement to old value instead of asserting
2906 A to be zero. Fixes wxhaskell-related crashes, and no other
2907 ill effects have been observed.
2909 Update: the reason why we're seeing these more elaborate
2910 relocations is due to a switch in how the NCG compiles SRTs
2911 and offsets to them from info tables. SRTs live in .(ro)data,
2912 while info tables live in .text, causing GAS to emit REL32/DISP32
2913 relocations with non-zero values. Adding the displacement is
2914 the right thing to do.
2916 *pP = S - ((UInt32)pP) - 4 + A;
2919 debugBelch("%s: unhandled PEi386 relocation type %d",
2920 oc->fileName, reltab_j->Type);
2927 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2931 #endif /* defined(OBJFORMAT_PEi386) */
2934 /* --------------------------------------------------------------------------
2936 * ------------------------------------------------------------------------*/
2938 #if defined(OBJFORMAT_ELF)
2943 #if defined(sparc_HOST_ARCH)
2944 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2945 #elif defined(i386_HOST_ARCH)
2946 # define ELF_TARGET_386 /* Used inside <elf.h> */
2947 #elif defined(x86_64_HOST_ARCH)
2948 # define ELF_TARGET_X64_64
2952 #if !defined(openbsd_HOST_OS)
2955 /* openbsd elf has things in different places, with diff names */
2956 # include <elf_abi.h>
2957 # include <machine/reloc.h>
2958 # define R_386_32 RELOC_32
2959 # define R_386_PC32 RELOC_PC32
2962 /* If elf.h doesn't define it */
2963 # ifndef R_X86_64_PC64
2964 # define R_X86_64_PC64 24
2968 * Define a set of types which can be used for both ELF32 and ELF64
2972 #define ELFCLASS ELFCLASS64
2973 #define Elf_Addr Elf64_Addr
2974 #define Elf_Word Elf64_Word
2975 #define Elf_Sword Elf64_Sword
2976 #define Elf_Ehdr Elf64_Ehdr
2977 #define Elf_Phdr Elf64_Phdr
2978 #define Elf_Shdr Elf64_Shdr
2979 #define Elf_Sym Elf64_Sym
2980 #define Elf_Rel Elf64_Rel
2981 #define Elf_Rela Elf64_Rela
2983 #define ELF_ST_TYPE ELF64_ST_TYPE
2986 #define ELF_ST_BIND ELF64_ST_BIND
2989 #define ELF_R_TYPE ELF64_R_TYPE
2992 #define ELF_R_SYM ELF64_R_SYM
2995 #define ELFCLASS ELFCLASS32
2996 #define Elf_Addr Elf32_Addr
2997 #define Elf_Word Elf32_Word
2998 #define Elf_Sword Elf32_Sword
2999 #define Elf_Ehdr Elf32_Ehdr
3000 #define Elf_Phdr Elf32_Phdr
3001 #define Elf_Shdr Elf32_Shdr
3002 #define Elf_Sym Elf32_Sym
3003 #define Elf_Rel Elf32_Rel
3004 #define Elf_Rela Elf32_Rela
3006 #define ELF_ST_TYPE ELF32_ST_TYPE
3009 #define ELF_ST_BIND ELF32_ST_BIND
3012 #define ELF_R_TYPE ELF32_R_TYPE
3015 #define ELF_R_SYM ELF32_R_SYM
3021 * Functions to allocate entries in dynamic sections. Currently we simply
3022 * preallocate a large number, and we don't check if a entry for the given
3023 * target already exists (a linear search is too slow). Ideally these
3024 * entries would be associated with symbols.
3027 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
3028 #define GOT_SIZE 0x20000
3029 #define FUNCTION_TABLE_SIZE 0x10000
3030 #define PLT_SIZE 0x08000
3033 static Elf_Addr got[GOT_SIZE];
3034 static unsigned int gotIndex;
3035 static Elf_Addr gp_val = (Elf_Addr)got;
3038 allocateGOTEntry(Elf_Addr target)
3042 if (gotIndex >= GOT_SIZE)
3043 barf("Global offset table overflow");
3045 entry = &got[gotIndex++];
3047 return (Elf_Addr)entry;
3051 #ifdef ELF_FUNCTION_DESC
3057 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
3058 static unsigned int functionTableIndex;
3061 allocateFunctionDesc(Elf_Addr target)
3063 FunctionDesc *entry;
3065 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
3066 barf("Function table overflow");
3068 entry = &functionTable[functionTableIndex++];
3070 entry->gp = (Elf_Addr)gp_val;
3071 return (Elf_Addr)entry;
3075 copyFunctionDesc(Elf_Addr target)
3077 FunctionDesc *olddesc = (FunctionDesc *)target;
3078 FunctionDesc *newdesc;
3080 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
3081 newdesc->gp = olddesc->gp;
3082 return (Elf_Addr)newdesc;
3089 unsigned char code[sizeof(plt_code)];
3093 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
3095 PLTEntry *plt = (PLTEntry *)oc->plt;
3098 if (oc->pltIndex >= PLT_SIZE)
3099 barf("Procedure table overflow");
3101 entry = &plt[oc->pltIndex++];
3102 memcpy(entry->code, plt_code, sizeof(entry->code));
3103 PLT_RELOC(entry->code, target);
3104 return (Elf_Addr)entry;
3110 return (PLT_SIZE * sizeof(PLTEntry));
3116 * Generic ELF functions
3120 findElfSection ( void* objImage, Elf_Word sh_type )
3122 char* ehdrC = (char*)objImage;
3123 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3124 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
3125 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3129 for (i = 0; i < ehdr->e_shnum; i++) {
3130 if (shdr[i].sh_type == sh_type
3131 /* Ignore the section header's string table. */
3132 && i != ehdr->e_shstrndx
3133 /* Ignore string tables named .stabstr, as they contain
3135 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3137 ptr = ehdrC + shdr[i].sh_offset;
3145 ocVerifyImage_ELF ( ObjectCode* oc )
3149 int i, j, nent, nstrtab, nsymtabs;
3153 char* ehdrC = (char*)(oc->image);
3154 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3156 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
3157 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
3158 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
3159 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
3160 errorBelch("%s: not an ELF object", oc->fileName);
3164 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
3165 errorBelch("%s: unsupported ELF format", oc->fileName);
3169 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
3170 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
3172 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
3173 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
3175 errorBelch("%s: unknown endiannness", oc->fileName);
3179 if (ehdr->e_type != ET_REL) {
3180 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
3183 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
3185 IF_DEBUG(linker,debugBelch( "Architecture is " ));
3186 switch (ehdr->e_machine) {
3187 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
3188 #ifdef EM_SPARC32PLUS
3189 case EM_SPARC32PLUS:
3191 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
3193 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
3195 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
3197 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
3198 #elif defined(EM_AMD64)
3199 case EM_AMD64: IF_DEBUG(linker,debugBelch( "amd64" )); break;
3201 default: IF_DEBUG(linker,debugBelch( "unknown" ));
3202 errorBelch("%s: unknown architecture (e_machine == %d)"
3203 , oc->fileName, ehdr->e_machine);
3207 IF_DEBUG(linker,debugBelch(
3208 "\nSection header table: start %ld, n_entries %d, ent_size %d\n",
3209 (long)ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
3211 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
3213 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3215 if (ehdr->e_shstrndx == SHN_UNDEF) {
3216 errorBelch("%s: no section header string table", oc->fileName);
3219 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
3221 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3224 for (i = 0; i < ehdr->e_shnum; i++) {
3225 IF_DEBUG(linker,debugBelch("%2d: ", i ));
3226 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
3227 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
3228 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
3229 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
3230 ehdrC + shdr[i].sh_offset,
3231 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
3233 if (shdr[i].sh_type == SHT_REL) {
3234 IF_DEBUG(linker,debugBelch("Rel " ));
3235 } else if (shdr[i].sh_type == SHT_RELA) {
3236 IF_DEBUG(linker,debugBelch("RelA " ));
3238 IF_DEBUG(linker,debugBelch(" "));
3241 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
3245 IF_DEBUG(linker,debugBelch( "\nString tables" ));
3248 for (i = 0; i < ehdr->e_shnum; i++) {
3249 if (shdr[i].sh_type == SHT_STRTAB
3250 /* Ignore the section header's string table. */
3251 && i != ehdr->e_shstrndx
3252 /* Ignore string tables named .stabstr, as they contain
3254 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3256 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
3257 strtab = ehdrC + shdr[i].sh_offset;
3262 errorBelch("%s: no string tables, or too many", oc->fileName);
3267 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
3268 for (i = 0; i < ehdr->e_shnum; i++) {
3269 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3270 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
3272 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3273 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3274 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%ld rem)\n",
3276 (long)shdr[i].sh_size % sizeof(Elf_Sym)
3278 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
3279 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
3282 for (j = 0; j < nent; j++) {
3283 IF_DEBUG(linker,debugBelch(" %2d ", j ));
3284 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
3285 (int)stab[j].st_shndx,
3286 (int)stab[j].st_size,
3287 (char*)stab[j].st_value ));
3289 IF_DEBUG(linker,debugBelch("type=" ));
3290 switch (ELF_ST_TYPE(stab[j].st_info)) {
3291 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
3292 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
3293 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
3294 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
3295 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
3296 default: IF_DEBUG(linker,debugBelch("? " )); break;
3298 IF_DEBUG(linker,debugBelch(" " ));
3300 IF_DEBUG(linker,debugBelch("bind=" ));
3301 switch (ELF_ST_BIND(stab[j].st_info)) {
3302 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
3303 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
3304 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
3305 default: IF_DEBUG(linker,debugBelch("? " )); break;
3307 IF_DEBUG(linker,debugBelch(" " ));
3309 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
3313 if (nsymtabs == 0) {
3314 errorBelch("%s: didn't find any symbol tables", oc->fileName);
3321 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
3325 if (hdr->sh_type == SHT_PROGBITS
3326 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
3327 /* .text-style section */
3328 return SECTIONKIND_CODE_OR_RODATA;
3331 if (hdr->sh_type == SHT_PROGBITS
3332 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3333 /* .data-style section */
3334 return SECTIONKIND_RWDATA;
3337 if (hdr->sh_type == SHT_PROGBITS
3338 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
3339 /* .rodata-style section */
3340 return SECTIONKIND_CODE_OR_RODATA;
3343 if (hdr->sh_type == SHT_NOBITS
3344 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3345 /* .bss-style section */
3347 return SECTIONKIND_RWDATA;
3350 return SECTIONKIND_OTHER;
3355 ocGetNames_ELF ( ObjectCode* oc )
3360 char* ehdrC = (char*)(oc->image);
3361 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3362 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
3363 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3365 ASSERT(symhash != NULL);
3368 errorBelch("%s: no strtab", oc->fileName);
3373 for (i = 0; i < ehdr->e_shnum; i++) {
3374 /* Figure out what kind of section it is. Logic derived from
3375 Figure 1.14 ("Special Sections") of the ELF document
3376 ("Portable Formats Specification, Version 1.1"). */
3378 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
3380 if (is_bss && shdr[i].sh_size > 0) {
3381 /* This is a non-empty .bss section. Allocate zeroed space for
3382 it, and set its .sh_offset field such that
3383 ehdrC + .sh_offset == addr_of_zeroed_space. */
3384 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
3385 "ocGetNames_ELF(BSS)");
3386 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
3388 debugBelch("BSS section at 0x%x, size %d\n",
3389 zspace, shdr[i].sh_size);
3393 /* fill in the section info */
3394 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
3395 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
3396 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
3397 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
3400 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3402 /* copy stuff into this module's object symbol table */
3403 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3404 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3406 oc->n_symbols = nent;
3407 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3408 "ocGetNames_ELF(oc->symbols)");
3410 for (j = 0; j < nent; j++) {
3412 char isLocal = FALSE; /* avoids uninit-var warning */
3414 char* nm = strtab + stab[j].st_name;
3415 int secno = stab[j].st_shndx;
3417 /* Figure out if we want to add it; if so, set ad to its
3418 address. Otherwise leave ad == NULL. */
3420 if (secno == SHN_COMMON) {
3422 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
3424 debugBelch("COMMON symbol, size %d name %s\n",
3425 stab[j].st_size, nm);
3427 /* Pointless to do addProddableBlock() for this area,
3428 since the linker should never poke around in it. */
3431 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
3432 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
3434 /* and not an undefined symbol */
3435 && stab[j].st_shndx != SHN_UNDEF
3436 /* and not in a "special section" */
3437 && stab[j].st_shndx < SHN_LORESERVE
3439 /* and it's a not a section or string table or anything silly */
3440 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
3441 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
3442 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
3445 /* Section 0 is the undefined section, hence > and not >=. */
3446 ASSERT(secno > 0 && secno < ehdr->e_shnum);
3448 if (shdr[secno].sh_type == SHT_NOBITS) {
3449 debugBelch(" BSS symbol, size %d off %d name %s\n",
3450 stab[j].st_size, stab[j].st_value, nm);
3453 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
3454 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
3457 #ifdef ELF_FUNCTION_DESC
3458 /* dlsym() and the initialisation table both give us function
3459 * descriptors, so to be consistent we store function descriptors
3460 * in the symbol table */
3461 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
3462 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
3464 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s\n",
3465 ad, oc->fileName, nm ));
3470 /* And the decision is ... */
3474 oc->symbols[j] = nm;
3477 /* Ignore entirely. */
3479 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
3483 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
3484 strtab + stab[j].st_name ));
3487 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
3488 (int)ELF_ST_BIND(stab[j].st_info),
3489 (int)ELF_ST_TYPE(stab[j].st_info),
3490 (int)stab[j].st_shndx,
3491 strtab + stab[j].st_name
3494 oc->symbols[j] = NULL;
3503 /* Do ELF relocations which lack an explicit addend. All x86-linux
3504 relocations appear to be of this form. */
3506 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
3507 Elf_Shdr* shdr, int shnum,
3508 Elf_Sym* stab, char* strtab )
3513 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
3514 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
3515 int target_shndx = shdr[shnum].sh_info;
3516 int symtab_shndx = shdr[shnum].sh_link;
3518 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3519 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
3520 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3521 target_shndx, symtab_shndx ));
3523 /* Skip sections that we're not interested in. */
3526 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
3527 if (kind == SECTIONKIND_OTHER) {
3528 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
3533 for (j = 0; j < nent; j++) {
3534 Elf_Addr offset = rtab[j].r_offset;
3535 Elf_Addr info = rtab[j].r_info;
3537 Elf_Addr P = ((Elf_Addr)targ) + offset;
3538 Elf_Word* pP = (Elf_Word*)P;
3543 StgStablePtr stablePtr;
3546 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
3547 j, (void*)offset, (void*)info ));
3549 IF_DEBUG(linker,debugBelch( " ZERO" ));
3552 Elf_Sym sym = stab[ELF_R_SYM(info)];
3553 /* First see if it is a local symbol. */
3554 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3555 /* Yes, so we can get the address directly from the ELF symbol
3557 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3559 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3560 + stab[ELF_R_SYM(info)].st_value);
3563 symbol = strtab + sym.st_name;
3564 stablePtr = (StgStablePtr)lookupHashTable(stablehash, (StgWord)symbol);
3565 if (NULL == stablePtr) {
3566 /* No, so look up the name in our global table. */
3567 S_tmp = lookupSymbol( symbol );
3568 S = (Elf_Addr)S_tmp;
3570 stableVal = deRefStablePtr( stablePtr );
3572 S = (Elf_Addr)S_tmp;
3576 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3579 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
3582 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
3583 (void*)P, (void*)S, (void*)A ));
3584 checkProddableBlock ( oc, pP );
3588 switch (ELF_R_TYPE(info)) {
3589 # ifdef i386_HOST_ARCH
3590 case R_386_32: *pP = value; break;
3591 case R_386_PC32: *pP = value - P; break;
3594 errorBelch("%s: unhandled ELF relocation(Rel) type %lu\n",
3595 oc->fileName, (lnat)ELF_R_TYPE(info));
3603 /* Do ELF relocations for which explicit addends are supplied.
3604 sparc-solaris relocations appear to be of this form. */
3606 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
3607 Elf_Shdr* shdr, int shnum,
3608 Elf_Sym* stab, char* strtab )
3611 char *symbol = NULL;
3613 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
3614 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
3615 int target_shndx = shdr[shnum].sh_info;
3616 int symtab_shndx = shdr[shnum].sh_link;
3618 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3619 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
3620 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3621 target_shndx, symtab_shndx ));
3623 for (j = 0; j < nent; j++) {
3624 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3625 /* This #ifdef only serves to avoid unused-var warnings. */
3626 Elf_Addr offset = rtab[j].r_offset;
3627 Elf_Addr P = targ + offset;
3629 Elf_Addr info = rtab[j].r_info;
3630 Elf_Addr A = rtab[j].r_addend;
3634 # if defined(sparc_HOST_ARCH)
3635 Elf_Word* pP = (Elf_Word*)P;
3637 # elif defined(powerpc_HOST_ARCH)
3641 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
3642 j, (void*)offset, (void*)info,
3645 IF_DEBUG(linker,debugBelch( " ZERO" ));
3648 Elf_Sym sym = stab[ELF_R_SYM(info)];
3649 /* First see if it is a local symbol. */
3650 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3651 /* Yes, so we can get the address directly from the ELF symbol
3653 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3655 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3656 + stab[ELF_R_SYM(info)].st_value);
3657 #ifdef ELF_FUNCTION_DESC
3658 /* Make a function descriptor for this function */
3659 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
3660 S = allocateFunctionDesc(S + A);
3665 /* No, so look up the name in our global table. */
3666 symbol = strtab + sym.st_name;
3667 S_tmp = lookupSymbol( symbol );
3668 S = (Elf_Addr)S_tmp;
3670 #ifdef ELF_FUNCTION_DESC
3671 /* If a function, already a function descriptor - we would
3672 have to copy it to add an offset. */
3673 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3674 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3678 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3681 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3684 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3685 (void*)P, (void*)S, (void*)A ));
3686 /* checkProddableBlock ( oc, (void*)P ); */
3690 switch (ELF_R_TYPE(info)) {
3691 # if defined(sparc_HOST_ARCH)
3692 case R_SPARC_WDISP30:
3693 w1 = *pP & 0xC0000000;
3694 w2 = (Elf_Word)((value - P) >> 2);
3695 ASSERT((w2 & 0xC0000000) == 0);
3700 w1 = *pP & 0xFFC00000;
3701 w2 = (Elf_Word)(value >> 10);
3702 ASSERT((w2 & 0xFFC00000) == 0);
3708 w2 = (Elf_Word)(value & 0x3FF);
3709 ASSERT((w2 & ~0x3FF) == 0);
3714 /* According to the Sun documentation:
3716 This relocation type resembles R_SPARC_32, except it refers to an
3717 unaligned word. That is, the word to be relocated must be treated
3718 as four separate bytes with arbitrary alignment, not as a word
3719 aligned according to the architecture requirements.
3722 w2 = (Elf_Word)value;
3724 // SPARC doesn't do misaligned writes of 32 bit words,
3725 // so we have to do this one byte-at-a-time.
3726 char *pPc = (char*)pP;
3727 pPc[0] = (char) ((Elf_Word)(w2 & 0xff000000) >> 24);
3728 pPc[1] = (char) ((Elf_Word)(w2 & 0x00ff0000) >> 16);
3729 pPc[2] = (char) ((Elf_Word)(w2 & 0x0000ff00) >> 8);
3730 pPc[3] = (char) ((Elf_Word)(w2 & 0x000000ff));
3734 w2 = (Elf_Word)value;
3737 # elif defined(powerpc_HOST_ARCH)
3738 case R_PPC_ADDR16_LO:
3739 *(Elf32_Half*) P = value;
3742 case R_PPC_ADDR16_HI:
3743 *(Elf32_Half*) P = value >> 16;
3746 case R_PPC_ADDR16_HA:
3747 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3751 *(Elf32_Word *) P = value;
3755 *(Elf32_Word *) P = value - P;
3761 if( delta << 6 >> 6 != delta )
3763 value = (Elf_Addr) (&makeSymbolExtra( oc, ELF_R_SYM(info), value )
3767 if( value == 0 || delta << 6 >> 6 != delta )
3769 barf( "Unable to make SymbolExtra for #%d",
3775 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3776 | (delta & 0x3fffffc);
3780 #if x86_64_HOST_ARCH
3782 *(Elf64_Xword *)P = value;
3787 #if defined(ALWAYS_PIC)
3788 barf("R_X86_64_PC32 relocation, but ALWAYS_PIC.");
3790 StgInt64 off = value - P;
3791 if (off >= 0x7fffffffL || off < -0x80000000L) {
3792 #if X86_64_ELF_NONPIC_HACK
3793 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3795 off = pltAddress + A - P;
3797 barf("R_X86_64_PC32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3798 symbol, off, oc->fileName );
3801 *(Elf64_Word *)P = (Elf64_Word)off;
3808 StgInt64 off = value - P;
3809 *(Elf64_Word *)P = (Elf64_Word)off;
3814 #if defined(ALWAYS_PIC)
3815 barf("R_X86_64_32 relocation, but ALWAYS_PIC.");
3817 if (value >= 0x7fffffffL) {
3818 #if X86_64_ELF_NONPIC_HACK
3819 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3821 value = pltAddress + A;
3823 barf("R_X86_64_32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3824 symbol, value, oc->fileName );
3827 *(Elf64_Word *)P = (Elf64_Word)value;
3832 #if defined(ALWAYS_PIC)
3833 barf("R_X86_64_32S relocation, but ALWAYS_PIC.");
3835 if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) {
3836 #if X86_64_ELF_NONPIC_HACK
3837 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3839 value = pltAddress + A;
3841 barf("R_X86_64_32S relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3842 symbol, value, oc->fileName );
3845 *(Elf64_Sword *)P = (Elf64_Sword)value;
3849 case R_X86_64_GOTPCREL:
3851 StgInt64 gotAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)->addr;
3852 StgInt64 off = gotAddress + A - P;
3853 *(Elf64_Word *)P = (Elf64_Word)off;
3857 case R_X86_64_PLT32:
3859 #if defined(ALWAYS_PIC)
3860 barf("R_X86_64_PLT32 relocation, but ALWAYS_PIC.");
3862 StgInt64 off = value - P;
3863 if (off >= 0x7fffffffL || off < -0x80000000L) {
3864 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3866 off = pltAddress + A - P;
3868 *(Elf64_Word *)P = (Elf64_Word)off;
3875 errorBelch("%s: unhandled ELF relocation(RelA) type %lu\n",
3876 oc->fileName, (lnat)ELF_R_TYPE(info));
3885 ocResolve_ELF ( ObjectCode* oc )
3889 Elf_Sym* stab = NULL;
3890 char* ehdrC = (char*)(oc->image);
3891 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3892 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3894 /* first find "the" symbol table */
3895 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3897 /* also go find the string table */
3898 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3900 if (stab == NULL || strtab == NULL) {
3901 errorBelch("%s: can't find string or symbol table", oc->fileName);
3905 /* Process the relocation sections. */
3906 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3907 if (shdr[shnum].sh_type == SHT_REL) {
3908 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3909 shnum, stab, strtab );
3913 if (shdr[shnum].sh_type == SHT_RELA) {
3914 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3915 shnum, stab, strtab );
3920 #if defined(powerpc_HOST_ARCH)
3921 ocFlushInstructionCache( oc );
3928 * PowerPC & X86_64 ELF specifics
3931 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3933 static int ocAllocateSymbolExtras_ELF( ObjectCode *oc )
3939 ehdr = (Elf_Ehdr *) oc->image;
3940 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3942 for( i = 0; i < ehdr->e_shnum; i++ )
3943 if( shdr[i].sh_type == SHT_SYMTAB )
3946 if( i == ehdr->e_shnum )
3948 errorBelch( "This ELF file contains no symtab" );
3952 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3954 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3955 (int) shdr[i].sh_entsize, (int) sizeof( Elf_Sym ) );
3960 return ocAllocateSymbolExtras( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3963 #endif /* powerpc */
3967 /* --------------------------------------------------------------------------
3969 * ------------------------------------------------------------------------*/
3971 #if defined(OBJFORMAT_MACHO)
3974 Support for MachO linking on Darwin/MacOS X
3975 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3977 I hereby formally apologize for the hackish nature of this code.
3978 Things that need to be done:
3979 *) implement ocVerifyImage_MachO
3980 *) add still more sanity checks.
3983 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
3984 #define mach_header mach_header_64
3985 #define segment_command segment_command_64
3986 #define section section_64
3987 #define nlist nlist_64
3990 #ifdef powerpc_HOST_ARCH
3991 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3993 struct mach_header *header = (struct mach_header *) oc->image;
3994 struct load_command *lc = (struct load_command *) (header + 1);
3997 for( i = 0; i < header->ncmds; i++ )
3999 if( lc->cmd == LC_SYMTAB )
4001 // Find out the first and last undefined external
4002 // symbol, so we don't have to allocate too many
4004 struct symtab_command *symLC = (struct symtab_command *) lc;
4005 unsigned min = symLC->nsyms, max = 0;
4006 struct nlist *nlist =
4007 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
4009 for(i=0;i<symLC->nsyms;i++)
4011 if(nlist[i].n_type & N_STAB)
4013 else if(nlist[i].n_type & N_EXT)
4015 if((nlist[i].n_type & N_TYPE) == N_UNDF
4016 && (nlist[i].n_value == 0))
4026 return ocAllocateSymbolExtras(oc, max - min + 1, min);
4031 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
4033 return ocAllocateSymbolExtras(oc,0,0);
4036 #ifdef x86_64_HOST_ARCH
4037 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
4039 struct mach_header *header = (struct mach_header *) oc->image;
4040 struct load_command *lc = (struct load_command *) (header + 1);
4043 for( i = 0; i < header->ncmds; i++ )
4045 if( lc->cmd == LC_SYMTAB )
4047 // Just allocate one entry for every symbol
4048 struct symtab_command *symLC = (struct symtab_command *) lc;
4050 return ocAllocateSymbolExtras(oc, symLC->nsyms, 0);
4053 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
4055 return ocAllocateSymbolExtras(oc,0,0);
4059 static int ocVerifyImage_MachO(ObjectCode* oc)
4061 char *image = (char*) oc->image;
4062 struct mach_header *header = (struct mach_header*) image;
4064 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
4065 if(header->magic != MH_MAGIC_64) {
4066 errorBelch("%s: Bad magic. Expected: %08x, got: %08x.\n",
4067 oc->fileName, MH_MAGIC_64, header->magic);
4071 if(header->magic != MH_MAGIC) {
4072 errorBelch("%s: Bad magic. Expected: %08x, got: %08x.\n",
4073 oc->fileName, MH_MAGIC, header->magic);
4077 // FIXME: do some more verifying here
4081 static int resolveImports(
4084 struct symtab_command *symLC,
4085 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
4086 unsigned long *indirectSyms,
4087 struct nlist *nlist)
4090 size_t itemSize = 4;
4093 int isJumpTable = 0;
4094 if(!strcmp(sect->sectname,"__jump_table"))
4098 ASSERT(sect->reserved2 == itemSize);
4102 for(i=0; i*itemSize < sect->size;i++)
4104 // according to otool, reserved1 contains the first index into the indirect symbol table
4105 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
4106 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4109 if((symbol->n_type & N_TYPE) == N_UNDF
4110 && (symbol->n_type & N_EXT) && (symbol->n_value != 0))
4111 addr = (void*) (symbol->n_value);
4113 addr = lookupSymbol(nm);
4116 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
4124 checkProddableBlock(oc,image + sect->offset + i*itemSize);
4125 *(image + sect->offset + i*itemSize) = 0xe9; // jmp
4126 *(unsigned*)(image + sect->offset + i*itemSize + 1)
4127 = (char*)addr - (image + sect->offset + i*itemSize + 5);
4132 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
4133 ((void**)(image + sect->offset))[i] = addr;
4140 static unsigned long relocateAddress(
4143 struct section* sections,
4144 unsigned long address)
4147 for(i = 0; i < nSections; i++)
4149 if(sections[i].addr <= address
4150 && address < sections[i].addr + sections[i].size)
4152 return (unsigned long)oc->image
4153 + sections[i].offset + address - sections[i].addr;
4156 barf("Invalid Mach-O file:"
4157 "Address out of bounds while relocating object file");
4161 static int relocateSection(
4164 struct symtab_command *symLC, struct nlist *nlist,
4165 int nSections, struct section* sections, struct section *sect)
4167 struct relocation_info *relocs;
4170 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
4172 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
4174 else if(!strcmp(sect->sectname,"__la_sym_ptr2"))
4176 else if(!strcmp(sect->sectname,"__la_sym_ptr3"))
4180 relocs = (struct relocation_info*) (image + sect->reloff);
4184 #ifdef x86_64_HOST_ARCH
4185 struct relocation_info *reloc = &relocs[i];
4187 char *thingPtr = image + sect->offset + reloc->r_address;
4189 /* We shouldn't need to initialise this, but gcc on OS X 64 bit
4190 complains that it may be used uninitialized if we don't */
4193 int type = reloc->r_type;
4195 checkProddableBlock(oc,thingPtr);
4196 switch(reloc->r_length)
4199 thing = *(uint8_t*)thingPtr;
4200 baseValue = (uint64_t)thingPtr + 1;
4203 thing = *(uint16_t*)thingPtr;
4204 baseValue = (uint64_t)thingPtr + 2;
4207 thing = *(uint32_t*)thingPtr;
4208 baseValue = (uint64_t)thingPtr + 4;
4211 thing = *(uint64_t*)thingPtr;
4212 baseValue = (uint64_t)thingPtr + 8;
4215 barf("Unknown size.");
4218 if(type == X86_64_RELOC_GOT
4219 || type == X86_64_RELOC_GOT_LOAD)
4221 ASSERT(reloc->r_extern);
4222 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)->addr;
4224 type = X86_64_RELOC_SIGNED;
4226 else if(reloc->r_extern)
4228 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4229 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4230 if(symbol->n_value == 0)
4231 value = (uint64_t) lookupSymbol(nm);
4233 value = relocateAddress(oc, nSections, sections,
4238 value = sections[reloc->r_symbolnum-1].offset
4239 - sections[reloc->r_symbolnum-1].addr
4243 if(type == X86_64_RELOC_BRANCH)
4245 if((int32_t)(value - baseValue) != (int64_t)(value - baseValue))
4247 ASSERT(reloc->r_extern);
4248 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)
4251 ASSERT((int32_t)(value - baseValue) == (int64_t)(value - baseValue));
4252 type = X86_64_RELOC_SIGNED;
4257 case X86_64_RELOC_UNSIGNED:
4258 ASSERT(!reloc->r_pcrel);
4261 case X86_64_RELOC_SIGNED:
4262 case X86_64_RELOC_SIGNED_1:
4263 case X86_64_RELOC_SIGNED_2:
4264 case X86_64_RELOC_SIGNED_4:
4265 ASSERT(reloc->r_pcrel);
4266 thing += value - baseValue;
4268 case X86_64_RELOC_SUBTRACTOR:
4269 ASSERT(!reloc->r_pcrel);
4273 barf("unkown relocation");
4276 switch(reloc->r_length)
4279 *(uint8_t*)thingPtr = thing;
4282 *(uint16_t*)thingPtr = thing;
4285 *(uint32_t*)thingPtr = thing;
4288 *(uint64_t*)thingPtr = thing;
4292 if(relocs[i].r_address & R_SCATTERED)
4294 struct scattered_relocation_info *scat =
4295 (struct scattered_relocation_info*) &relocs[i];
4299 if(scat->r_length == 2)
4301 unsigned long word = 0;
4302 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
4303 checkProddableBlock(oc,wordPtr);
4305 // Note on relocation types:
4306 // i386 uses the GENERIC_RELOC_* types,
4307 // while ppc uses special PPC_RELOC_* types.
4308 // *_RELOC_VANILLA and *_RELOC_PAIR have the same value
4309 // in both cases, all others are different.
4310 // Therefore, we use GENERIC_RELOC_VANILLA
4311 // and GENERIC_RELOC_PAIR instead of the PPC variants,
4312 // and use #ifdefs for the other types.
4314 // Step 1: Figure out what the relocated value should be
4315 if(scat->r_type == GENERIC_RELOC_VANILLA)
4317 word = *wordPtr + (unsigned long) relocateAddress(
4324 #ifdef powerpc_HOST_ARCH
4325 else if(scat->r_type == PPC_RELOC_SECTDIFF
4326 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
4327 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
4328 || scat->r_type == PPC_RELOC_HA16_SECTDIFF
4329 || scat->r_type == PPC_RELOC_LOCAL_SECTDIFF)
4331 else if(scat->r_type == GENERIC_RELOC_SECTDIFF
4332 || scat->r_type == GENERIC_RELOC_LOCAL_SECTDIFF)
4335 struct scattered_relocation_info *pair =
4336 (struct scattered_relocation_info*) &relocs[i+1];
4338 if(!pair->r_scattered || pair->r_type != GENERIC_RELOC_PAIR)
4339 barf("Invalid Mach-O file: "
4340 "RELOC_*_SECTDIFF not followed by RELOC_PAIR");
4342 word = (unsigned long)
4343 (relocateAddress(oc, nSections, sections, scat->r_value)
4344 - relocateAddress(oc, nSections, sections, pair->r_value));
4347 #ifdef powerpc_HOST_ARCH
4348 else if(scat->r_type == PPC_RELOC_HI16
4349 || scat->r_type == PPC_RELOC_LO16
4350 || scat->r_type == PPC_RELOC_HA16
4351 || scat->r_type == PPC_RELOC_LO14)
4352 { // these are generated by label+offset things
4353 struct relocation_info *pair = &relocs[i+1];
4354 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
4355 barf("Invalid Mach-O file: "
4356 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
4358 if(scat->r_type == PPC_RELOC_LO16)
4360 word = ((unsigned short*) wordPtr)[1];
4361 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4363 else if(scat->r_type == PPC_RELOC_LO14)
4365 barf("Unsupported Relocation: PPC_RELOC_LO14");
4366 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
4367 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4369 else if(scat->r_type == PPC_RELOC_HI16)
4371 word = ((unsigned short*) wordPtr)[1] << 16;
4372 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4374 else if(scat->r_type == PPC_RELOC_HA16)
4376 word = ((unsigned short*) wordPtr)[1] << 16;
4377 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4381 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
4389 barf ("Don't know how to handle this Mach-O "
4390 "scattered relocation entry: "
4391 "object file %s; entry type %ld; "
4393 oc->fileName, scat->r_type, scat->r_address);
4397 #ifdef powerpc_HOST_ARCH
4398 if(scat->r_type == GENERIC_RELOC_VANILLA
4399 || scat->r_type == PPC_RELOC_SECTDIFF)
4401 if(scat->r_type == GENERIC_RELOC_VANILLA
4402 || scat->r_type == GENERIC_RELOC_SECTDIFF
4403 || scat->r_type == GENERIC_RELOC_LOCAL_SECTDIFF)
4408 #ifdef powerpc_HOST_ARCH
4409 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
4411 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4413 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
4415 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4417 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
4419 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4420 + ((word & (1<<15)) ? 1 : 0);
4426 barf("Can't handle Mach-O scattered relocation entry "
4427 "with this r_length tag: "
4428 "object file %s; entry type %ld; "
4429 "r_length tag %ld; address %#lx\n",
4430 oc->fileName, scat->r_type, scat->r_length,
4435 else /* scat->r_pcrel */
4437 barf("Don't know how to handle *PC-relative* Mach-O "
4438 "scattered relocation entry: "
4439 "object file %s; entry type %ld; address %#lx\n",
4440 oc->fileName, scat->r_type, scat->r_address);
4445 else /* !(relocs[i].r_address & R_SCATTERED) */
4447 struct relocation_info *reloc = &relocs[i];
4448 if(reloc->r_pcrel && !reloc->r_extern)
4451 if(reloc->r_length == 2)
4453 unsigned long word = 0;
4454 #ifdef powerpc_HOST_ARCH
4455 unsigned long jumpIsland = 0;
4456 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
4457 // to avoid warning and to catch
4461 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
4462 checkProddableBlock(oc,wordPtr);
4464 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4468 #ifdef powerpc_HOST_ARCH
4469 else if(reloc->r_type == PPC_RELOC_LO16)
4471 word = ((unsigned short*) wordPtr)[1];
4472 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4474 else if(reloc->r_type == PPC_RELOC_HI16)
4476 word = ((unsigned short*) wordPtr)[1] << 16;
4477 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4479 else if(reloc->r_type == PPC_RELOC_HA16)
4481 word = ((unsigned short*) wordPtr)[1] << 16;
4482 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4484 else if(reloc->r_type == PPC_RELOC_BR24)
4487 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
4492 barf("Can't handle this Mach-O relocation entry "
4494 "object file %s; entry type %ld; address %#lx\n",
4495 oc->fileName, reloc->r_type, reloc->r_address);
4499 if(!reloc->r_extern)
4502 sections[reloc->r_symbolnum-1].offset
4503 - sections[reloc->r_symbolnum-1].addr
4510 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4511 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4512 void *symbolAddress = lookupSymbol(nm);
4515 errorBelch("\nunknown symbol `%s'", nm);
4521 #ifdef powerpc_HOST_ARCH
4522 // In the .o file, this should be a relative jump to NULL
4523 // and we'll change it to a relative jump to the symbol
4524 ASSERT(word + reloc->r_address == 0);
4525 jumpIsland = (unsigned long)
4526 &makeSymbolExtra(oc,
4528 (unsigned long) symbolAddress)
4532 offsetToJumpIsland = word + jumpIsland
4533 - (((long)image) + sect->offset - sect->addr);
4536 word += (unsigned long) symbolAddress
4537 - (((long)image) + sect->offset - sect->addr);
4541 word += (unsigned long) symbolAddress;
4545 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4550 #ifdef powerpc_HOST_ARCH
4551 else if(reloc->r_type == PPC_RELOC_LO16)
4553 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4556 else if(reloc->r_type == PPC_RELOC_HI16)
4558 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4561 else if(reloc->r_type == PPC_RELOC_HA16)
4563 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4564 + ((word & (1<<15)) ? 1 : 0);
4567 else if(reloc->r_type == PPC_RELOC_BR24)
4569 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4571 // The branch offset is too large.
4572 // Therefore, we try to use a jump island.
4575 barf("unconditional relative branch out of range: "
4576 "no jump island available");
4579 word = offsetToJumpIsland;
4580 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4581 barf("unconditional relative branch out of range: "
4582 "jump island out of range");
4584 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
4591 barf("Can't handle Mach-O relocation entry (not scattered) "
4592 "with this r_length tag: "
4593 "object file %s; entry type %ld; "
4594 "r_length tag %ld; address %#lx\n",
4595 oc->fileName, reloc->r_type, reloc->r_length,
4605 static int ocGetNames_MachO(ObjectCode* oc)
4607 char *image = (char*) oc->image;
4608 struct mach_header *header = (struct mach_header*) image;
4609 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4610 unsigned i,curSymbol = 0;
4611 struct segment_command *segLC = NULL;
4612 struct section *sections;
4613 struct symtab_command *symLC = NULL;
4614 struct nlist *nlist;
4615 unsigned long commonSize = 0;
4616 char *commonStorage = NULL;
4617 unsigned long commonCounter;
4619 for(i=0;i<header->ncmds;i++)
4621 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4622 segLC = (struct segment_command*) lc;
4623 else if(lc->cmd == LC_SYMTAB)
4624 symLC = (struct symtab_command*) lc;
4625 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4628 sections = (struct section*) (segLC+1);
4629 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4633 barf("ocGetNames_MachO: no segment load command");
4635 for(i=0;i<segLC->nsects;i++)
4637 if(sections[i].size == 0)
4640 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
4642 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
4643 "ocGetNames_MachO(common symbols)");
4644 sections[i].offset = zeroFillArea - image;
4647 if(!strcmp(sections[i].sectname,"__text"))
4648 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
4649 (void*) (image + sections[i].offset),
4650 (void*) (image + sections[i].offset + sections[i].size));
4651 else if(!strcmp(sections[i].sectname,"__const"))
4652 addSection(oc, SECTIONKIND_RWDATA,
4653 (void*) (image + sections[i].offset),
4654 (void*) (image + sections[i].offset + sections[i].size));
4655 else if(!strcmp(sections[i].sectname,"__data"))
4656 addSection(oc, SECTIONKIND_RWDATA,
4657 (void*) (image + sections[i].offset),
4658 (void*) (image + sections[i].offset + sections[i].size));
4659 else if(!strcmp(sections[i].sectname,"__bss")
4660 || !strcmp(sections[i].sectname,"__common"))
4661 addSection(oc, SECTIONKIND_RWDATA,
4662 (void*) (image + sections[i].offset),
4663 (void*) (image + sections[i].offset + sections[i].size));
4665 addProddableBlock(oc, (void*) (image + sections[i].offset),
4669 // count external symbols defined here
4673 for(i=0;i<symLC->nsyms;i++)
4675 if(nlist[i].n_type & N_STAB)
4677 else if(nlist[i].n_type & N_EXT)
4679 if((nlist[i].n_type & N_TYPE) == N_UNDF
4680 && (nlist[i].n_value != 0))
4682 commonSize += nlist[i].n_value;
4685 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4690 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
4691 "ocGetNames_MachO(oc->symbols)");
4695 for(i=0;i<symLC->nsyms;i++)
4697 if(nlist[i].n_type & N_STAB)
4699 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4701 if(nlist[i].n_type & N_EXT)
4703 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4704 if((nlist[i].n_desc & N_WEAK_DEF) && lookupSymbol(nm))
4705 ; // weak definition, and we already have a definition
4708 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4710 + sections[nlist[i].n_sect-1].offset
4711 - sections[nlist[i].n_sect-1].addr
4712 + nlist[i].n_value);
4713 oc->symbols[curSymbol++] = nm;
4720 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
4721 commonCounter = (unsigned long)commonStorage;
4724 for(i=0;i<symLC->nsyms;i++)
4726 if((nlist[i].n_type & N_TYPE) == N_UNDF
4727 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
4729 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4730 unsigned long sz = nlist[i].n_value;
4732 nlist[i].n_value = commonCounter;
4734 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4735 (void*)commonCounter);
4736 oc->symbols[curSymbol++] = nm;
4738 commonCounter += sz;
4745 static int ocResolve_MachO(ObjectCode* oc)
4747 char *image = (char*) oc->image;
4748 struct mach_header *header = (struct mach_header*) image;
4749 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4751 struct segment_command *segLC = NULL;
4752 struct section *sections;
4753 struct symtab_command *symLC = NULL;
4754 struct dysymtab_command *dsymLC = NULL;
4755 struct nlist *nlist;
4757 for(i=0;i<header->ncmds;i++)
4759 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4760 segLC = (struct segment_command*) lc;
4761 else if(lc->cmd == LC_SYMTAB)
4762 symLC = (struct symtab_command*) lc;
4763 else if(lc->cmd == LC_DYSYMTAB)
4764 dsymLC = (struct dysymtab_command*) lc;
4765 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4768 sections = (struct section*) (segLC+1);
4769 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4774 unsigned long *indirectSyms
4775 = (unsigned long*) (image + dsymLC->indirectsymoff);
4777 for(i=0;i<segLC->nsects;i++)
4779 if( !strcmp(sections[i].sectname,"__la_symbol_ptr")
4780 || !strcmp(sections[i].sectname,"__la_sym_ptr2")
4781 || !strcmp(sections[i].sectname,"__la_sym_ptr3"))
4783 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4786 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr")
4787 || !strcmp(sections[i].sectname,"__pointers"))
4789 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4792 else if(!strcmp(sections[i].sectname,"__jump_table"))
4794 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4800 for(i=0;i<segLC->nsects;i++)
4802 if(!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
4806 #if defined (powerpc_HOST_ARCH)
4807 ocFlushInstructionCache( oc );
4813 #ifdef powerpc_HOST_ARCH
4815 * The Mach-O object format uses leading underscores. But not everywhere.
4816 * There is a small number of runtime support functions defined in
4817 * libcc_dynamic.a whose name does not have a leading underscore.
4818 * As a consequence, we can't get their address from C code.
4819 * We have to use inline assembler just to take the address of a function.
4823 extern void* symbolsWithoutUnderscore[];
4825 static void machoInitSymbolsWithoutUnderscore()
4827 void **p = symbolsWithoutUnderscore;
4828 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
4830 #undef SymI_NeedsProto
4831 #define SymI_NeedsProto(x) \
4832 __asm__ volatile(".long " # x);
4834 RTS_MACHO_NOUNDERLINE_SYMBOLS
4836 __asm__ volatile(".text");
4838 #undef SymI_NeedsProto
4839 #define SymI_NeedsProto(x) \
4840 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
4842 RTS_MACHO_NOUNDERLINE_SYMBOLS
4844 #undef SymI_NeedsProto
4850 * Figure out by how much to shift the entire Mach-O file in memory
4851 * when loading so that its single segment ends up 16-byte-aligned
4853 static int machoGetMisalignment( FILE * f )
4855 struct mach_header header;
4858 fread(&header, sizeof(header), 1, f);
4861 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
4862 if(header.magic != MH_MAGIC_64) {
4863 errorBelch("Bad magic. Expected: %08x, got: %08x.\n",
4864 MH_MAGIC_64, header->magic);
4868 if(header.magic != MH_MAGIC) {
4869 errorBelch("Bad magic. Expected: %08x, got: %08x.\n",
4870 MH_MAGIC, header->magic);
4875 misalignment = (header.sizeofcmds + sizeof(header))
4878 return misalignment ? (16 - misalignment) : 0;