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(setIOManagerPipe) \
523 SymI_HasProto(ioManagerWakeup) \
524 SymI_HasProto(ioManagerSync) \
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(getOrSetGHCConcSignalHandlerStore) \
774 SymI_HasProto(getOrSetGHCConcPendingEventsStore) \
775 SymI_HasProto(getOrSetGHCConcPendingDelaysStore) \
776 SymI_HasProto(getOrSetGHCConcIOManagerThreadStore) \
777 SymI_HasProto(getOrSetGHCConcProddingStore) \
778 SymI_HasProto(genSymZh) \
779 SymI_HasProto(genericRaise) \
780 SymI_HasProto(getProgArgv) \
781 SymI_HasProto(getFullProgArgv) \
782 SymI_HasProto(getStablePtr) \
783 SymI_HasProto(hs_init) \
784 SymI_HasProto(hs_exit) \
785 SymI_HasProto(hs_set_argv) \
786 SymI_HasProto(hs_add_root) \
787 SymI_HasProto(hs_perform_gc) \
788 SymI_HasProto(hs_free_stable_ptr) \
789 SymI_HasProto(hs_free_fun_ptr) \
790 SymI_HasProto(hs_hpc_rootModule) \
791 SymI_HasProto(hs_hpc_module) \
792 SymI_HasProto(initLinker) \
793 SymI_HasProto(stg_unpackClosurezh) \
794 SymI_HasProto(stg_getApStackValzh) \
795 SymI_HasProto(stg_getSparkzh) \
796 SymI_HasProto(stg_numSparkszh) \
797 SymI_HasProto(stg_isCurrentThreadBoundzh) \
798 SymI_HasProto(stg_isEmptyMVarzh) \
799 SymI_HasProto(stg_killThreadzh) \
800 SymI_HasProto(loadObj) \
801 SymI_HasProto(insertStableSymbol) \
802 SymI_HasProto(insertSymbol) \
803 SymI_HasProto(lookupSymbol) \
804 SymI_HasProto(stg_makeStablePtrzh) \
805 SymI_HasProto(stg_mkApUpd0zh) \
806 SymI_HasProto(stg_myThreadIdzh) \
807 SymI_HasProto(stg_labelThreadzh) \
808 SymI_HasProto(stg_newArrayzh) \
809 SymI_HasProto(stg_newBCOzh) \
810 SymI_HasProto(stg_newByteArrayzh) \
811 SymI_HasProto_redirect(newCAF, newDynCAF) \
812 SymI_HasProto(stg_newMVarzh) \
813 SymI_HasProto(stg_newMutVarzh) \
814 SymI_HasProto(stg_newTVarzh) \
815 SymI_HasProto(stg_noDuplicatezh) \
816 SymI_HasProto(stg_atomicModifyMutVarzh) \
817 SymI_HasProto(stg_newPinnedByteArrayzh) \
818 SymI_HasProto(stg_newAlignedPinnedByteArrayzh) \
819 SymI_HasProto(newSpark) \
820 SymI_HasProto(performGC) \
821 SymI_HasProto(performMajorGC) \
822 SymI_HasProto(prog_argc) \
823 SymI_HasProto(prog_argv) \
824 SymI_HasProto(stg_putMVarzh) \
825 SymI_HasProto(stg_raisezh) \
826 SymI_HasProto(stg_raiseIOzh) \
827 SymI_HasProto(stg_readTVarzh) \
828 SymI_HasProto(stg_readTVarIOzh) \
829 SymI_HasProto(resumeThread) \
830 SymI_HasProto(resolveObjs) \
831 SymI_HasProto(stg_retryzh) \
832 SymI_HasProto(rts_apply) \
833 SymI_HasProto(rts_checkSchedStatus) \
834 SymI_HasProto(rts_eval) \
835 SymI_HasProto(rts_evalIO) \
836 SymI_HasProto(rts_evalLazyIO) \
837 SymI_HasProto(rts_evalStableIO) \
838 SymI_HasProto(rts_eval_) \
839 SymI_HasProto(rts_getBool) \
840 SymI_HasProto(rts_getChar) \
841 SymI_HasProto(rts_getDouble) \
842 SymI_HasProto(rts_getFloat) \
843 SymI_HasProto(rts_getInt) \
844 SymI_HasProto(rts_getInt8) \
845 SymI_HasProto(rts_getInt16) \
846 SymI_HasProto(rts_getInt32) \
847 SymI_HasProto(rts_getInt64) \
848 SymI_HasProto(rts_getPtr) \
849 SymI_HasProto(rts_getFunPtr) \
850 SymI_HasProto(rts_getStablePtr) \
851 SymI_HasProto(rts_getThreadId) \
852 SymI_HasProto(rts_getWord) \
853 SymI_HasProto(rts_getWord8) \
854 SymI_HasProto(rts_getWord16) \
855 SymI_HasProto(rts_getWord32) \
856 SymI_HasProto(rts_getWord64) \
857 SymI_HasProto(rts_lock) \
858 SymI_HasProto(rts_mkBool) \
859 SymI_HasProto(rts_mkChar) \
860 SymI_HasProto(rts_mkDouble) \
861 SymI_HasProto(rts_mkFloat) \
862 SymI_HasProto(rts_mkInt) \
863 SymI_HasProto(rts_mkInt8) \
864 SymI_HasProto(rts_mkInt16) \
865 SymI_HasProto(rts_mkInt32) \
866 SymI_HasProto(rts_mkInt64) \
867 SymI_HasProto(rts_mkPtr) \
868 SymI_HasProto(rts_mkFunPtr) \
869 SymI_HasProto(rts_mkStablePtr) \
870 SymI_HasProto(rts_mkString) \
871 SymI_HasProto(rts_mkWord) \
872 SymI_HasProto(rts_mkWord8) \
873 SymI_HasProto(rts_mkWord16) \
874 SymI_HasProto(rts_mkWord32) \
875 SymI_HasProto(rts_mkWord64) \
876 SymI_HasProto(rts_unlock) \
877 SymI_HasProto(rts_unsafeGetMyCapability) \
878 SymI_HasProto(rtsSupportsBoundThreads) \
879 SymI_HasProto(setProgArgv) \
880 SymI_HasProto(startupHaskell) \
881 SymI_HasProto(shutdownHaskell) \
882 SymI_HasProto(shutdownHaskellAndExit) \
883 SymI_HasProto(stable_ptr_table) \
884 SymI_HasProto(stackOverflow) \
885 SymI_HasProto(stg_CAF_BLACKHOLE_info) \
886 SymI_HasProto(stg_BLACKHOLE_info) \
887 SymI_HasProto(__stg_EAGER_BLACKHOLE_info) \
888 SymI_HasProto(stg_BLOCKING_QUEUE_CLEAN_info) \
889 SymI_HasProto(stg_BLOCKING_QUEUE_DIRTY_info) \
890 SymI_HasProto(startTimer) \
891 SymI_HasProto(stg_MVAR_CLEAN_info) \
892 SymI_HasProto(stg_MVAR_DIRTY_info) \
893 SymI_HasProto(stg_IND_STATIC_info) \
894 SymI_HasProto(stg_ARR_WORDS_info) \
895 SymI_HasProto(stg_MUT_ARR_PTRS_DIRTY_info) \
896 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN_info) \
897 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN0_info) \
898 SymI_HasProto(stg_WEAK_info) \
899 SymI_HasProto(stg_ap_v_info) \
900 SymI_HasProto(stg_ap_f_info) \
901 SymI_HasProto(stg_ap_d_info) \
902 SymI_HasProto(stg_ap_l_info) \
903 SymI_HasProto(stg_ap_n_info) \
904 SymI_HasProto(stg_ap_p_info) \
905 SymI_HasProto(stg_ap_pv_info) \
906 SymI_HasProto(stg_ap_pp_info) \
907 SymI_HasProto(stg_ap_ppv_info) \
908 SymI_HasProto(stg_ap_ppp_info) \
909 SymI_HasProto(stg_ap_pppv_info) \
910 SymI_HasProto(stg_ap_pppp_info) \
911 SymI_HasProto(stg_ap_ppppp_info) \
912 SymI_HasProto(stg_ap_pppppp_info) \
913 SymI_HasProto(stg_ap_0_fast) \
914 SymI_HasProto(stg_ap_v_fast) \
915 SymI_HasProto(stg_ap_f_fast) \
916 SymI_HasProto(stg_ap_d_fast) \
917 SymI_HasProto(stg_ap_l_fast) \
918 SymI_HasProto(stg_ap_n_fast) \
919 SymI_HasProto(stg_ap_p_fast) \
920 SymI_HasProto(stg_ap_pv_fast) \
921 SymI_HasProto(stg_ap_pp_fast) \
922 SymI_HasProto(stg_ap_ppv_fast) \
923 SymI_HasProto(stg_ap_ppp_fast) \
924 SymI_HasProto(stg_ap_pppv_fast) \
925 SymI_HasProto(stg_ap_pppp_fast) \
926 SymI_HasProto(stg_ap_ppppp_fast) \
927 SymI_HasProto(stg_ap_pppppp_fast) \
928 SymI_HasProto(stg_ap_1_upd_info) \
929 SymI_HasProto(stg_ap_2_upd_info) \
930 SymI_HasProto(stg_ap_3_upd_info) \
931 SymI_HasProto(stg_ap_4_upd_info) \
932 SymI_HasProto(stg_ap_5_upd_info) \
933 SymI_HasProto(stg_ap_6_upd_info) \
934 SymI_HasProto(stg_ap_7_upd_info) \
935 SymI_HasProto(stg_exit) \
936 SymI_HasProto(stg_sel_0_upd_info) \
937 SymI_HasProto(stg_sel_10_upd_info) \
938 SymI_HasProto(stg_sel_11_upd_info) \
939 SymI_HasProto(stg_sel_12_upd_info) \
940 SymI_HasProto(stg_sel_13_upd_info) \
941 SymI_HasProto(stg_sel_14_upd_info) \
942 SymI_HasProto(stg_sel_15_upd_info) \
943 SymI_HasProto(stg_sel_1_upd_info) \
944 SymI_HasProto(stg_sel_2_upd_info) \
945 SymI_HasProto(stg_sel_3_upd_info) \
946 SymI_HasProto(stg_sel_4_upd_info) \
947 SymI_HasProto(stg_sel_5_upd_info) \
948 SymI_HasProto(stg_sel_6_upd_info) \
949 SymI_HasProto(stg_sel_7_upd_info) \
950 SymI_HasProto(stg_sel_8_upd_info) \
951 SymI_HasProto(stg_sel_9_upd_info) \
952 SymI_HasProto(stg_upd_frame_info) \
953 SymI_HasProto(stg_bh_upd_frame_info) \
954 SymI_HasProto(suspendThread) \
955 SymI_HasProto(stg_takeMVarzh) \
956 SymI_HasProto(stg_threadStatuszh) \
957 SymI_HasProto(stg_tryPutMVarzh) \
958 SymI_HasProto(stg_tryTakeMVarzh) \
959 SymI_HasProto(stg_unmaskAsyncExceptionszh) \
960 SymI_HasProto(unloadObj) \
961 SymI_HasProto(stg_unsafeThawArrayzh) \
962 SymI_HasProto(stg_waitReadzh) \
963 SymI_HasProto(stg_waitWritezh) \
964 SymI_HasProto(stg_writeTVarzh) \
965 SymI_HasProto(stg_yieldzh) \
966 SymI_NeedsProto(stg_interp_constr_entry) \
967 SymI_HasProto(alloc_blocks_lim) \
969 SymI_HasProto(allocate) \
970 SymI_HasProto(allocateExec) \
971 SymI_HasProto(freeExec) \
972 SymI_HasProto(getAllocations) \
973 SymI_HasProto(revertCAFs) \
974 SymI_HasProto(RtsFlags) \
975 SymI_NeedsProto(rts_breakpoint_io_action) \
976 SymI_NeedsProto(rts_stop_next_breakpoint) \
977 SymI_NeedsProto(rts_stop_on_exception) \
978 SymI_HasProto(stopTimer) \
979 SymI_HasProto(n_capabilities) \
980 SymI_HasProto(stg_traceCcszh) \
981 SymI_HasProto(stg_traceEventzh) \
982 RTS_USER_SIGNALS_SYMBOLS \
986 // 64-bit support functions in libgcc.a
987 #if defined(__GNUC__) && SIZEOF_VOID_P <= 4
988 #define RTS_LIBGCC_SYMBOLS \
989 SymI_NeedsProto(__divdi3) \
990 SymI_NeedsProto(__udivdi3) \
991 SymI_NeedsProto(__moddi3) \
992 SymI_NeedsProto(__umoddi3) \
993 SymI_NeedsProto(__muldi3) \
994 SymI_NeedsProto(__ashldi3) \
995 SymI_NeedsProto(__ashrdi3) \
996 SymI_NeedsProto(__lshrdi3)
998 #define RTS_LIBGCC_SYMBOLS
1001 #if defined(darwin_HOST_OS) && defined(powerpc_HOST_ARCH)
1002 // Symbols that don't have a leading underscore
1003 // on Mac OS X. They have to receive special treatment,
1004 // see machoInitSymbolsWithoutUnderscore()
1005 #define RTS_MACHO_NOUNDERLINE_SYMBOLS \
1006 SymI_NeedsProto(saveFP) \
1007 SymI_NeedsProto(restFP)
1010 /* entirely bogus claims about types of these symbols */
1011 #define SymI_NeedsProto(vvv) extern void vvv(void);
1012 #if defined(__PIC__) && defined(mingw32_HOST_OS)
1013 #define SymE_HasProto(vvv) SymE_HasProto(vvv);
1014 #define SymE_NeedsProto(vvv) extern void _imp__ ## vvv (void);
1016 #define SymE_NeedsProto(vvv) SymI_NeedsProto(vvv);
1017 #define SymE_HasProto(vvv) SymI_HasProto(vvv)
1019 #define SymI_HasProto(vvv) /**/
1020 #define SymI_HasProto_redirect(vvv,xxx) /**/
1023 RTS_POSIX_ONLY_SYMBOLS
1024 RTS_MINGW_ONLY_SYMBOLS
1025 RTS_CYGWIN_ONLY_SYMBOLS
1026 RTS_DARWIN_ONLY_SYMBOLS
1029 #undef SymI_NeedsProto
1030 #undef SymI_HasProto
1031 #undef SymI_HasProto_redirect
1032 #undef SymE_HasProto
1033 #undef SymE_NeedsProto
1035 #ifdef LEADING_UNDERSCORE
1036 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
1038 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
1041 #define SymI_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1043 #define SymE_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1044 (void*)DLL_IMPORT_DATA_REF(vvv) },
1046 #define SymI_NeedsProto(vvv) SymI_HasProto(vvv)
1047 #define SymE_NeedsProto(vvv) SymE_HasProto(vvv)
1049 // SymI_HasProto_redirect allows us to redirect references to one symbol to
1050 // another symbol. See newCAF/newDynCAF for an example.
1051 #define SymI_HasProto_redirect(vvv,xxx) \
1052 { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1055 static RtsSymbolVal rtsSyms[] = {
1058 RTS_POSIX_ONLY_SYMBOLS
1059 RTS_MINGW_ONLY_SYMBOLS
1060 RTS_CYGWIN_ONLY_SYMBOLS
1061 RTS_DARWIN_ONLY_SYMBOLS
1064 #if defined(darwin_HOST_OS) && defined(i386_HOST_ARCH)
1065 // dyld stub code contains references to this,
1066 // but it should never be called because we treat
1067 // lazy pointers as nonlazy.
1068 { "dyld_stub_binding_helper", (void*)0xDEADBEEF },
1070 { 0, 0 } /* sentinel */
1075 /* -----------------------------------------------------------------------------
1076 * Insert symbols into hash tables, checking for duplicates.
1079 static void ghciInsertStrHashTable ( char* obj_name,
1085 if (lookupHashTable(table, (StgWord)key) == NULL)
1087 insertStrHashTable(table, (StgWord)key, data);
1092 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
1094 "whilst processing object file\n"
1096 "This could be caused by:\n"
1097 " * Loading two different object files which export the same symbol\n"
1098 " * Specifying the same object file twice on the GHCi command line\n"
1099 " * An incorrect `package.conf' entry, causing some object to be\n"
1101 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
1108 /* -----------------------------------------------------------------------------
1109 * initialize the object linker
1113 static int linker_init_done = 0 ;
1115 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1116 static void *dl_prog_handle;
1117 static regex_t re_invalid;
1118 static regex_t re_realso;
1120 static Mutex dl_mutex; // mutex to protect dlopen/dlerror critical section
1128 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1132 /* Make initLinker idempotent, so we can call it
1133 before evey relevant operation; that means we
1134 don't need to initialise the linker separately */
1135 if (linker_init_done == 1) { return; } else {
1136 linker_init_done = 1;
1139 #if defined(THREADED_RTS) && (defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO))
1140 initMutex(&dl_mutex);
1142 stablehash = allocStrHashTable();
1143 symhash = allocStrHashTable();
1145 /* populate the symbol table with stuff from the RTS */
1146 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
1147 ghciInsertStrHashTable("(GHCi built-in symbols)",
1148 symhash, sym->lbl, sym->addr);
1150 # if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH)
1151 machoInitSymbolsWithoutUnderscore();
1154 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1155 # if defined(RTLD_DEFAULT)
1156 dl_prog_handle = RTLD_DEFAULT;
1158 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
1159 # endif /* RTLD_DEFAULT */
1161 compileResult = regcomp(&re_invalid,
1162 "(([^ \t()])+\\.so([^ \t:()])*):([ \t])*invalid ELF header",
1164 ASSERT( compileResult == 0 );
1165 compileResult = regcomp(&re_realso,
1166 "GROUP *\\( *(([^ )])+)",
1168 ASSERT( compileResult == 0 );
1171 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
1172 if (RtsFlags.MiscFlags.linkerMemBase != 0) {
1173 // User-override for mmap_32bit_base
1174 mmap_32bit_base = (void*)RtsFlags.MiscFlags.linkerMemBase;
1178 #if defined(mingw32_HOST_OS)
1180 * These two libraries cause problems when added to the static link,
1181 * but are necessary for resolving symbols in GHCi, hence we load
1182 * them manually here.
1190 exitLinker( void ) {
1191 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1192 if (linker_init_done == 1) {
1193 regfree(&re_invalid);
1194 regfree(&re_realso);
1196 closeMutex(&dl_mutex);
1202 /* -----------------------------------------------------------------------------
1203 * Loading DLL or .so dynamic libraries
1204 * -----------------------------------------------------------------------------
1206 * Add a DLL from which symbols may be found. In the ELF case, just
1207 * do RTLD_GLOBAL-style add, so no further messing around needs to
1208 * happen in order that symbols in the loaded .so are findable --
1209 * lookupSymbol() will subsequently see them by dlsym on the program's
1210 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
1212 * In the PEi386 case, open the DLLs and put handles to them in a
1213 * linked list. When looking for a symbol, try all handles in the
1214 * list. This means that we need to load even DLLs that are guaranteed
1215 * to be in the ghc.exe image already, just so we can get a handle
1216 * to give to loadSymbol, so that we can find the symbols. For such
1217 * libraries, the LoadLibrary call should be a no-op except for returning
1222 #if defined(OBJFORMAT_PEi386)
1223 /* A record for storing handles into DLLs. */
1228 struct _OpenedDLL* next;
1233 /* A list thereof. */
1234 static OpenedDLL* opened_dlls = NULL;
1237 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1240 internal_dlopen(const char *dll_name)
1246 // omitted: RTLD_NOW
1247 // see http://www.haskell.org/pipermail/cvs-ghc/2007-September/038570.html
1249 debugBelch("internal_dlopen: dll_name = '%s'\n", dll_name));
1251 //-------------- Begin critical section ------------------
1252 // This critical section is necessary because dlerror() is not
1253 // required to be reentrant (see POSIX -- IEEE Std 1003.1-2008)
1254 // Also, the error message returned must be copied to preserve it
1257 ACQUIRE_LOCK(&dl_mutex);
1258 hdl = dlopen(dll_name, RTLD_LAZY | RTLD_GLOBAL);
1262 /* dlopen failed; return a ptr to the error msg. */
1264 if (errmsg == NULL) errmsg = "addDLL: unknown error";
1265 errmsg_copy = stgMallocBytes(strlen(errmsg)+1, "addDLL");
1266 strcpy(errmsg_copy, errmsg);
1267 errmsg = errmsg_copy;
1269 RELEASE_LOCK(&dl_mutex);
1270 //--------------- End critical section -------------------
1277 addDLL( char *dll_name )
1279 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1280 /* ------------------- ELF DLL loader ------------------- */
1283 regmatch_t match[NMATCH];
1286 size_t match_length;
1287 #define MAXLINE 1000
1293 IF_DEBUG(linker, debugBelch("addDLL: dll_name = '%s'\n", dll_name));
1294 errmsg = internal_dlopen(dll_name);
1296 if (errmsg == NULL) {
1300 // GHC Trac ticket #2615
1301 // On some systems (e.g., Gentoo Linux) dynamic files (e.g. libc.so)
1302 // contain linker scripts rather than ELF-format object code. This
1303 // code handles the situation by recognizing the real object code
1304 // file name given in the linker script.
1306 // If an "invalid ELF header" error occurs, it is assumed that the
1307 // .so file contains a linker script instead of ELF object code.
1308 // In this case, the code looks for the GROUP ( ... ) linker
1309 // directive. If one is found, the first file name inside the
1310 // parentheses is treated as the name of a dynamic library and the
1311 // code attempts to dlopen that file. If this is also unsuccessful,
1312 // an error message is returned.
1314 // see if the error message is due to an invalid ELF header
1315 IF_DEBUG(linker, debugBelch("errmsg = '%s'\n", errmsg));
1316 result = regexec(&re_invalid, errmsg, (size_t) NMATCH, match, 0);
1317 IF_DEBUG(linker, debugBelch("result = %i\n", result));
1319 // success -- try to read the named file as a linker script
1320 match_length = (size_t) stg_min((match[1].rm_eo - match[1].rm_so),
1322 strncpy(line, (errmsg+(match[1].rm_so)),match_length);
1323 line[match_length] = '\0'; // make sure string is null-terminated
1324 IF_DEBUG(linker, debugBelch ("file name = '%s'\n", line));
1325 if ((fp = fopen(line, "r")) == NULL) {
1326 return errmsg; // return original error if open fails
1328 // try to find a GROUP ( ... ) command
1329 while (fgets(line, MAXLINE, fp) != NULL) {
1330 IF_DEBUG(linker, debugBelch("input line = %s", line));
1331 if (regexec(&re_realso, line, (size_t) NMATCH, match, 0) == 0) {
1332 // success -- try to dlopen the first named file
1333 IF_DEBUG(linker, debugBelch("match%s\n",""));
1334 line[match[1].rm_eo] = '\0';
1335 errmsg = internal_dlopen(line+match[1].rm_so);
1338 // if control reaches here, no GROUP ( ... ) directive was found
1339 // and the original error message is returned to the caller
1345 # elif defined(OBJFORMAT_PEi386)
1346 /* ------------------- Win32 DLL loader ------------------- */
1354 /* debugBelch("\naddDLL; dll_name = `%s'\n", dll_name); */
1356 /* See if we've already got it, and ignore if so. */
1357 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
1358 if (0 == strcmp(o_dll->name, dll_name))
1362 /* The file name has no suffix (yet) so that we can try
1363 both foo.dll and foo.drv
1365 The documentation for LoadLibrary says:
1366 If no file name extension is specified in the lpFileName
1367 parameter, the default library extension .dll is
1368 appended. However, the file name string can include a trailing
1369 point character (.) to indicate that the module name has no
1372 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
1373 sprintf(buf, "%s.DLL", dll_name);
1374 instance = LoadLibrary(buf);
1375 if (instance == NULL) {
1376 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1377 // KAA: allow loading of drivers (like winspool.drv)
1378 sprintf(buf, "%s.DRV", dll_name);
1379 instance = LoadLibrary(buf);
1380 if (instance == NULL) {
1381 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1382 // #1883: allow loading of unix-style libfoo.dll DLLs
1383 sprintf(buf, "lib%s.DLL", dll_name);
1384 instance = LoadLibrary(buf);
1385 if (instance == NULL) {
1392 /* Add this DLL to the list of DLLs in which to search for symbols. */
1393 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
1394 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
1395 strcpy(o_dll->name, dll_name);
1396 o_dll->instance = instance;
1397 o_dll->next = opened_dlls;
1398 opened_dlls = o_dll;
1404 sysErrorBelch(dll_name);
1406 /* LoadLibrary failed; return a ptr to the error msg. */
1407 return "addDLL: could not load DLL";
1410 barf("addDLL: not implemented on this platform");
1414 /* -----------------------------------------------------------------------------
1415 * insert a stable symbol in the hash table
1419 insertStableSymbol(char* obj_name, char* key, StgPtr p)
1421 ghciInsertStrHashTable(obj_name, stablehash, key, getStablePtr(p));
1425 /* -----------------------------------------------------------------------------
1426 * insert a symbol in the hash table
1429 insertSymbol(char* obj_name, char* key, void* data)
1431 ghciInsertStrHashTable(obj_name, symhash, key, data);
1434 /* -----------------------------------------------------------------------------
1435 * lookup a symbol in the hash table
1438 lookupSymbol( char *lbl )
1442 ASSERT(symhash != NULL);
1443 val = lookupStrHashTable(symhash, lbl);
1446 # if defined(OBJFORMAT_ELF)
1447 return dlsym(dl_prog_handle, lbl);
1448 # elif defined(OBJFORMAT_MACHO)
1450 /* On OS X 10.3 and later, we use dlsym instead of the old legacy
1453 HACK: On OS X, global symbols are prefixed with an underscore.
1454 However, dlsym wants us to omit the leading underscore from the
1455 symbol name. For now, we simply strip it off here (and ONLY
1458 ASSERT(lbl[0] == '_');
1459 return dlsym(dl_prog_handle, lbl+1);
1461 if(NSIsSymbolNameDefined(lbl)) {
1462 NSSymbol symbol = NSLookupAndBindSymbol(lbl);
1463 return NSAddressOfSymbol(symbol);
1467 # endif /* HAVE_DLFCN_H */
1468 # elif defined(OBJFORMAT_PEi386)
1471 sym = lookupSymbolInDLLs((unsigned char*)lbl);
1472 if (sym != NULL) { return sym; };
1474 // Also try looking up the symbol without the @N suffix. Some
1475 // DLLs have the suffixes on their symbols, some don't.
1476 zapTrailingAtSign ( (unsigned char*)lbl );
1477 sym = lookupSymbolInDLLs((unsigned char*)lbl);
1478 if (sym != NULL) { return sym; };
1490 /* -----------------------------------------------------------------------------
1491 * Debugging aid: look in GHCi's object symbol tables for symbols
1492 * within DELTA bytes of the specified address, and show their names.
1495 void ghci_enquire ( char* addr );
1497 void ghci_enquire ( char* addr )
1502 const int DELTA = 64;
1507 for (oc = objects; oc; oc = oc->next) {
1508 for (i = 0; i < oc->n_symbols; i++) {
1509 sym = oc->symbols[i];
1510 if (sym == NULL) continue;
1513 a = lookupStrHashTable(symhash, sym);
1516 // debugBelch("ghci_enquire: can't find %s\n", sym);
1518 else if (addr-DELTA <= a && a <= addr+DELTA) {
1519 debugBelch("%p + %3d == `%s'\n", addr, (int)(a - addr), sym);
1527 #define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))
1530 mmapForLinker (size_t bytes, nat flags, int fd)
1532 void *map_addr = NULL;
1535 static nat fixed = 0;
1537 pagesize = getpagesize();
1538 size = ROUND_UP(bytes, pagesize);
1540 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
1543 if (mmap_32bit_base != 0) {
1544 map_addr = mmap_32bit_base;
1548 result = mmap(map_addr, size, PROT_EXEC|PROT_READ|PROT_WRITE,
1549 MAP_PRIVATE|TRY_MAP_32BIT|fixed|flags, fd, 0);
1551 if (result == MAP_FAILED) {
1552 sysErrorBelch("mmap %lu bytes at %p",(lnat)size,map_addr);
1553 errorBelch("Try specifying an address with +RTS -xm<addr> -RTS");
1554 stg_exit(EXIT_FAILURE);
1557 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
1558 if (mmap_32bit_base != 0) {
1559 if (result == map_addr) {
1560 mmap_32bit_base = (StgWord8*)map_addr + size;
1562 if ((W_)result > 0x80000000) {
1563 // oops, we were given memory over 2Gb
1564 #if defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS)
1565 // Some platforms require MAP_FIXED. This is normally
1566 // a bad idea, because MAP_FIXED will overwrite
1567 // existing mappings.
1568 munmap(result,size);
1572 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);
1575 // hmm, we were given memory somewhere else, but it's
1576 // still under 2Gb so we can use it. Next time, ask
1577 // for memory right after the place we just got some
1578 mmap_32bit_base = (StgWord8*)result + size;
1582 if ((W_)result > 0x80000000) {
1583 // oops, we were given memory over 2Gb
1584 // ... try allocating memory somewhere else?;
1585 debugTrace(DEBUG_linker,"MAP_32BIT didn't work; gave us %lu bytes at 0x%p", bytes, result);
1586 munmap(result, size);
1588 // Set a base address and try again... (guess: 1Gb)
1589 mmap_32bit_base = (void*)0x40000000;
1599 /* -----------------------------------------------------------------------------
1600 * Load an obj (populate the global symbol table, but don't resolve yet)
1602 * Returns: 1 if ok, 0 on error.
1605 loadObj( char *path )
1615 IF_DEBUG(linker, debugBelch("loadObj %s\n", path));
1618 /* debugBelch("loadObj %s\n", path ); */
1620 /* Check that we haven't already loaded this object.
1621 Ignore requests to load multiple times */
1625 for (o = objects; o; o = o->next) {
1626 if (0 == strcmp(o->fileName, path)) {
1628 break; /* don't need to search further */
1632 IF_DEBUG(linker, debugBelch(
1633 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1634 "same object file twice:\n"
1636 "GHCi will ignore this, but be warned.\n"
1638 return 1; /* success */
1642 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
1644 # if defined(OBJFORMAT_ELF)
1645 oc->formatName = "ELF";
1646 # elif defined(OBJFORMAT_PEi386)
1647 oc->formatName = "PEi386";
1648 # elif defined(OBJFORMAT_MACHO)
1649 oc->formatName = "Mach-O";
1652 barf("loadObj: not implemented on this platform");
1655 r = stat(path, &st);
1657 IF_DEBUG(linker, debugBelch("File doesn't exist\n"));
1661 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1662 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1663 strcpy(oc->fileName, path);
1665 oc->fileSize = st.st_size;
1667 oc->sections = NULL;
1668 oc->proddables = NULL;
1670 /* chain it onto the list of objects */
1675 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1677 #if defined(openbsd_HOST_OS)
1678 fd = open(path, O_RDONLY, S_IRUSR);
1680 fd = open(path, O_RDONLY);
1683 barf("loadObj: can't open `%s'", path);
1685 oc->image = mmapForLinker(oc->fileSize, 0, fd);
1689 #else /* !USE_MMAP */
1690 /* load the image into memory */
1691 f = fopen(path, "rb");
1693 barf("loadObj: can't read `%s'", path);
1695 # if defined(mingw32_HOST_OS)
1696 // TODO: We would like to use allocateExec here, but allocateExec
1697 // cannot currently allocate blocks large enough.
1698 oc->image = VirtualAlloc(NULL, oc->fileSize, MEM_RESERVE | MEM_COMMIT,
1699 PAGE_EXECUTE_READWRITE);
1700 # elif defined(darwin_HOST_OS)
1701 // In a Mach-O .o file, all sections can and will be misaligned
1702 // if the total size of the headers is not a multiple of the
1703 // desired alignment. This is fine for .o files that only serve
1704 // as input for the static linker, but it's not fine for us,
1705 // as SSE (used by gcc for floating point) and Altivec require
1706 // 16-byte alignment.
1707 // We calculate the correct alignment from the header before
1708 // reading the file, and then we misalign oc->image on purpose so
1709 // that the actual sections end up aligned again.
1710 oc->misalignment = machoGetMisalignment(f);
1711 oc->image = stgMallocBytes(oc->fileSize + oc->misalignment, "loadObj(image)");
1712 oc->image += oc->misalignment;
1714 oc->image = stgMallocBytes(oc->fileSize, "loadObj(image)");
1719 n = fread ( oc->image, 1, oc->fileSize, f );
1720 if (n != oc->fileSize)
1721 barf("loadObj: error whilst reading `%s'", path);
1724 #endif /* USE_MMAP */
1726 # if defined(OBJFORMAT_MACHO) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1727 r = ocAllocateSymbolExtras_MachO ( oc );
1729 IF_DEBUG(linker, debugBelch("ocAllocateSymbolExtras_MachO failed\n"));
1732 # elif defined(OBJFORMAT_ELF) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1733 r = ocAllocateSymbolExtras_ELF ( oc );
1735 IF_DEBUG(linker, debugBelch("ocAllocateSymbolExtras_ELF failed\n"));
1740 /* verify the in-memory image */
1741 # if defined(OBJFORMAT_ELF)
1742 r = ocVerifyImage_ELF ( oc );
1743 # elif defined(OBJFORMAT_PEi386)
1744 r = ocVerifyImage_PEi386 ( oc );
1745 # elif defined(OBJFORMAT_MACHO)
1746 r = ocVerifyImage_MachO ( oc );
1748 barf("loadObj: no verify method");
1751 IF_DEBUG(linker, debugBelch("ocVerifyImage_* failed\n"));
1755 /* build the symbol list for this image */
1756 # if defined(OBJFORMAT_ELF)
1757 r = ocGetNames_ELF ( oc );
1758 # elif defined(OBJFORMAT_PEi386)
1759 r = ocGetNames_PEi386 ( oc );
1760 # elif defined(OBJFORMAT_MACHO)
1761 r = ocGetNames_MachO ( oc );
1763 barf("loadObj: no getNames method");
1766 IF_DEBUG(linker, debugBelch("ocGetNames_* failed\n"));
1770 /* loaded, but not resolved yet */
1771 oc->status = OBJECT_LOADED;
1776 /* -----------------------------------------------------------------------------
1777 * resolve all the currently unlinked objects in memory
1779 * Returns: 1 if ok, 0 on error.
1789 for (oc = objects; oc; oc = oc->next) {
1790 if (oc->status != OBJECT_RESOLVED) {
1791 # if defined(OBJFORMAT_ELF)
1792 r = ocResolve_ELF ( oc );
1793 # elif defined(OBJFORMAT_PEi386)
1794 r = ocResolve_PEi386 ( oc );
1795 # elif defined(OBJFORMAT_MACHO)
1796 r = ocResolve_MachO ( oc );
1798 barf("resolveObjs: not implemented on this platform");
1800 if (!r) { return r; }
1801 oc->status = OBJECT_RESOLVED;
1807 /* -----------------------------------------------------------------------------
1808 * delete an object from the pool
1811 unloadObj( char *path )
1813 ObjectCode *oc, *prev;
1815 ASSERT(symhash != NULL);
1816 ASSERT(objects != NULL);
1821 for (oc = objects; oc; prev = oc, oc = oc->next) {
1822 if (!strcmp(oc->fileName,path)) {
1824 /* Remove all the mappings for the symbols within this
1829 for (i = 0; i < oc->n_symbols; i++) {
1830 if (oc->symbols[i] != NULL) {
1831 removeStrHashTable(symhash, oc->symbols[i], NULL);
1839 prev->next = oc->next;
1842 // We're going to leave this in place, in case there are
1843 // any pointers from the heap into it:
1844 // #ifdef mingw32_HOST_OS
1845 // VirtualFree(oc->image);
1847 // stgFree(oc->image);
1849 stgFree(oc->fileName);
1850 stgFree(oc->symbols);
1851 stgFree(oc->sections);
1857 errorBelch("unloadObj: can't find `%s' to unload", path);
1861 /* -----------------------------------------------------------------------------
1862 * Sanity checking. For each ObjectCode, maintain a list of address ranges
1863 * which may be prodded during relocation, and abort if we try and write
1864 * outside any of these.
1866 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
1869 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
1870 /* debugBelch("aPB %p %p %d\n", oc, start, size); */
1874 pb->next = oc->proddables;
1875 oc->proddables = pb;
1878 static void checkProddableBlock ( ObjectCode* oc, void* addr )
1881 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
1882 char* s = (char*)(pb->start);
1883 char* e = s + pb->size - 1;
1884 char* a = (char*)addr;
1885 /* Assumes that the biggest fixup involves a 4-byte write. This
1886 probably needs to be changed to 8 (ie, +7) on 64-bit
1888 if (a >= s && (a+3) <= e) return;
1890 barf("checkProddableBlock: invalid fixup in runtime linker");
1893 /* -----------------------------------------------------------------------------
1894 * Section management.
1896 static void addSection ( ObjectCode* oc, SectionKind kind,
1897 void* start, void* end )
1899 Section* s = stgMallocBytes(sizeof(Section), "addSection");
1903 s->next = oc->sections;
1906 debugBelch("addSection: %p-%p (size %d), kind %d\n",
1907 start, ((char*)end)-1, end - start + 1, kind );
1912 /* --------------------------------------------------------------------------
1914 * This is about allocating a small chunk of memory for every symbol in the
1915 * object file. We make sure that the SymboLExtras are always "in range" of
1916 * limited-range PC-relative instructions on various platforms by allocating
1917 * them right next to the object code itself.
1920 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
1923 ocAllocateSymbolExtras
1925 Allocate additional space at the end of the object file image to make room
1926 for jump islands (powerpc, x86_64) and GOT entries (x86_64).
1928 PowerPC relative branch instructions have a 24 bit displacement field.
1929 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
1930 If a particular imported symbol is outside this range, we have to redirect
1931 the jump to a short piece of new code that just loads the 32bit absolute
1932 address and jumps there.
1933 On x86_64, PC-relative jumps and PC-relative accesses to the GOT are limited
1936 This function just allocates space for one SymbolExtra for every
1937 undefined symbol in the object file. The code for the jump islands is
1938 filled in by makeSymbolExtra below.
1941 static int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first )
1948 int misalignment = 0;
1949 #ifdef darwin_HOST_OS
1950 misalignment = oc->misalignment;
1956 // round up to the nearest 4
1957 aligned = (oc->fileSize + 3) & ~3;
1960 pagesize = getpagesize();
1961 n = ROUND_UP( oc->fileSize, pagesize );
1962 m = ROUND_UP( aligned + sizeof (SymbolExtra) * count, pagesize );
1964 /* we try to use spare space at the end of the last page of the
1965 * image for the jump islands, but if there isn't enough space
1966 * then we have to map some (anonymously, remembering MAP_32BIT).
1968 if( m > n ) // we need to allocate more pages
1970 oc->symbol_extras = mmapForLinker(sizeof(SymbolExtra) * count,
1975 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1978 oc->image -= misalignment;
1979 oc->image = stgReallocBytes( oc->image,
1981 aligned + sizeof (SymbolExtra) * count,
1982 "ocAllocateSymbolExtras" );
1983 oc->image += misalignment;
1985 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1986 #endif /* USE_MMAP */
1988 memset( oc->symbol_extras, 0, sizeof (SymbolExtra) * count );
1991 oc->symbol_extras = NULL;
1993 oc->first_symbol_extra = first;
1994 oc->n_symbol_extras = count;
1999 static SymbolExtra* makeSymbolExtra( ObjectCode* oc,
2000 unsigned long symbolNumber,
2001 unsigned long target )
2005 ASSERT( symbolNumber >= oc->first_symbol_extra
2006 && symbolNumber - oc->first_symbol_extra < oc->n_symbol_extras);
2008 extra = &oc->symbol_extras[symbolNumber - oc->first_symbol_extra];
2010 #ifdef powerpc_HOST_ARCH
2011 // lis r12, hi16(target)
2012 extra->jumpIsland.lis_r12 = 0x3d80;
2013 extra->jumpIsland.hi_addr = target >> 16;
2015 // ori r12, r12, lo16(target)
2016 extra->jumpIsland.ori_r12_r12 = 0x618c;
2017 extra->jumpIsland.lo_addr = target & 0xffff;
2020 extra->jumpIsland.mtctr_r12 = 0x7d8903a6;
2023 extra->jumpIsland.bctr = 0x4e800420;
2025 #ifdef x86_64_HOST_ARCH
2027 static uint8_t jmp[] = { 0xFF, 0x25, 0xF2, 0xFF, 0xFF, 0xFF };
2028 extra->addr = target;
2029 memcpy(extra->jumpIsland, jmp, 6);
2037 /* --------------------------------------------------------------------------
2038 * PowerPC specifics (instruction cache flushing)
2039 * ------------------------------------------------------------------------*/
2041 #ifdef powerpc_HOST_ARCH
2043 ocFlushInstructionCache
2045 Flush the data & instruction caches.
2046 Because the PPC has split data/instruction caches, we have to
2047 do that whenever we modify code at runtime.
2050 static void ocFlushInstructionCache( ObjectCode *oc )
2052 int n = (oc->fileSize + sizeof( SymbolExtra ) * oc->n_symbol_extras + 3) / 4;
2053 unsigned long *p = (unsigned long *) oc->image;
2057 __asm__ volatile ( "dcbf 0,%0\n\t"
2065 __asm__ volatile ( "sync\n\t"
2071 /* --------------------------------------------------------------------------
2072 * PEi386 specifics (Win32 targets)
2073 * ------------------------------------------------------------------------*/
2075 /* The information for this linker comes from
2076 Microsoft Portable Executable
2077 and Common Object File Format Specification
2078 revision 5.1 January 1998
2079 which SimonM says comes from the MS Developer Network CDs.
2081 It can be found there (on older CDs), but can also be found
2084 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
2086 (this is Rev 6.0 from February 1999).
2088 Things move, so if that fails, try searching for it via
2090 http://www.google.com/search?q=PE+COFF+specification
2092 The ultimate reference for the PE format is the Winnt.h
2093 header file that comes with the Platform SDKs; as always,
2094 implementations will drift wrt their documentation.
2096 A good background article on the PE format is Matt Pietrek's
2097 March 1994 article in Microsoft System Journal (MSJ)
2098 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
2099 Win32 Portable Executable File Format." The info in there
2100 has recently been updated in a two part article in
2101 MSDN magazine, issues Feb and March 2002,
2102 "Inside Windows: An In-Depth Look into the Win32 Portable
2103 Executable File Format"
2105 John Levine's book "Linkers and Loaders" contains useful
2110 #if defined(OBJFORMAT_PEi386)
2114 typedef unsigned char UChar;
2115 typedef unsigned short UInt16;
2116 typedef unsigned int UInt32;
2123 UInt16 NumberOfSections;
2124 UInt32 TimeDateStamp;
2125 UInt32 PointerToSymbolTable;
2126 UInt32 NumberOfSymbols;
2127 UInt16 SizeOfOptionalHeader;
2128 UInt16 Characteristics;
2132 #define sizeof_COFF_header 20
2139 UInt32 VirtualAddress;
2140 UInt32 SizeOfRawData;
2141 UInt32 PointerToRawData;
2142 UInt32 PointerToRelocations;
2143 UInt32 PointerToLinenumbers;
2144 UInt16 NumberOfRelocations;
2145 UInt16 NumberOfLineNumbers;
2146 UInt32 Characteristics;
2150 #define sizeof_COFF_section 40
2157 UInt16 SectionNumber;
2160 UChar NumberOfAuxSymbols;
2164 #define sizeof_COFF_symbol 18
2169 UInt32 VirtualAddress;
2170 UInt32 SymbolTableIndex;
2175 #define sizeof_COFF_reloc 10
2178 /* From PE spec doc, section 3.3.2 */
2179 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
2180 windows.h -- for the same purpose, but I want to know what I'm
2182 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
2183 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
2184 #define MYIMAGE_FILE_DLL 0x2000
2185 #define MYIMAGE_FILE_SYSTEM 0x1000
2186 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
2187 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
2188 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
2190 /* From PE spec doc, section 5.4.2 and 5.4.4 */
2191 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
2192 #define MYIMAGE_SYM_CLASS_STATIC 3
2193 #define MYIMAGE_SYM_UNDEFINED 0
2195 /* From PE spec doc, section 4.1 */
2196 #define MYIMAGE_SCN_CNT_CODE 0x00000020
2197 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
2198 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
2200 /* From PE spec doc, section 5.2.1 */
2201 #define MYIMAGE_REL_I386_DIR32 0x0006
2202 #define MYIMAGE_REL_I386_REL32 0x0014
2205 /* We use myindex to calculate array addresses, rather than
2206 simply doing the normal subscript thing. That's because
2207 some of the above structs have sizes which are not
2208 a whole number of words. GCC rounds their sizes up to a
2209 whole number of words, which means that the address calcs
2210 arising from using normal C indexing or pointer arithmetic
2211 are just plain wrong. Sigh.
2214 myindex ( int scale, void* base, int index )
2217 ((UChar*)base) + scale * index;
2222 printName ( UChar* name, UChar* strtab )
2224 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2225 UInt32 strtab_offset = * (UInt32*)(name+4);
2226 debugBelch("%s", strtab + strtab_offset );
2229 for (i = 0; i < 8; i++) {
2230 if (name[i] == 0) break;
2231 debugBelch("%c", name[i] );
2238 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
2240 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2241 UInt32 strtab_offset = * (UInt32*)(name+4);
2242 strncpy ( (char*)dst, (char*)strtab+strtab_offset, dstSize );
2248 if (name[i] == 0) break;
2258 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
2261 /* If the string is longer than 8 bytes, look in the
2262 string table for it -- this will be correctly zero terminated.
2264 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2265 UInt32 strtab_offset = * (UInt32*)(name+4);
2266 return ((UChar*)strtab) + strtab_offset;
2268 /* Otherwise, if shorter than 8 bytes, return the original,
2269 which by defn is correctly terminated.
2271 if (name[7]==0) return name;
2272 /* The annoying case: 8 bytes. Copy into a temporary
2273 (which is never freed ...)
2275 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
2277 strncpy((char*)newstr,(char*)name,8);
2283 /* Just compares the short names (first 8 chars) */
2284 static COFF_section *
2285 findPEi386SectionCalled ( ObjectCode* oc, UChar* name )
2289 = (COFF_header*)(oc->image);
2290 COFF_section* sectab
2292 ((UChar*)(oc->image))
2293 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2295 for (i = 0; i < hdr->NumberOfSections; i++) {
2298 COFF_section* section_i
2300 myindex ( sizeof_COFF_section, sectab, i );
2301 n1 = (UChar*) &(section_i->Name);
2303 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
2304 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
2305 n1[6]==n2[6] && n1[7]==n2[7])
2314 zapTrailingAtSign ( UChar* sym )
2316 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
2318 if (sym[0] == 0) return;
2320 while (sym[i] != 0) i++;
2323 while (j > 0 && my_isdigit(sym[j])) j--;
2324 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
2329 lookupSymbolInDLLs ( UChar *lbl )
2334 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
2335 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
2337 if (lbl[0] == '_') {
2338 /* HACK: if the name has an initial underscore, try stripping
2339 it off & look that up first. I've yet to verify whether there's
2340 a Rule that governs whether an initial '_' *should always* be
2341 stripped off when mapping from import lib name to the DLL name.
2343 sym = GetProcAddress(o_dll->instance, (char*)(lbl+1));
2345 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
2349 sym = GetProcAddress(o_dll->instance, (char*)lbl);
2351 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
2360 ocVerifyImage_PEi386 ( ObjectCode* oc )
2365 COFF_section* sectab;
2366 COFF_symbol* symtab;
2368 /* debugBelch("\nLOADING %s\n", oc->fileName); */
2369 hdr = (COFF_header*)(oc->image);
2370 sectab = (COFF_section*) (
2371 ((UChar*)(oc->image))
2372 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2374 symtab = (COFF_symbol*) (
2375 ((UChar*)(oc->image))
2376 + hdr->PointerToSymbolTable
2378 strtab = ((UChar*)symtab)
2379 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2381 if (hdr->Machine != 0x14c) {
2382 errorBelch("%s: Not x86 PEi386", oc->fileName);
2385 if (hdr->SizeOfOptionalHeader != 0) {
2386 errorBelch("%s: PEi386 with nonempty optional header", oc->fileName);
2389 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
2390 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
2391 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
2392 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
2393 errorBelch("%s: Not a PEi386 object file", oc->fileName);
2396 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
2397 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
2398 errorBelch("%s: Invalid PEi386 word size or endiannness: %d",
2400 (int)(hdr->Characteristics));
2403 /* If the string table size is way crazy, this might indicate that
2404 there are more than 64k relocations, despite claims to the
2405 contrary. Hence this test. */
2406 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
2408 if ( (*(UInt32*)strtab) > 600000 ) {
2409 /* Note that 600k has no special significance other than being
2410 big enough to handle the almost-2MB-sized lumps that
2411 constitute HSwin32*.o. */
2412 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
2417 /* No further verification after this point; only debug printing. */
2419 IF_DEBUG(linker, i=1);
2420 if (i == 0) return 1;
2422 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
2423 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
2424 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
2427 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
2428 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
2429 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
2430 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
2431 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
2432 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
2433 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
2435 /* Print the section table. */
2437 for (i = 0; i < hdr->NumberOfSections; i++) {
2439 COFF_section* sectab_i
2441 myindex ( sizeof_COFF_section, sectab, i );
2448 printName ( sectab_i->Name, strtab );
2458 sectab_i->VirtualSize,
2459 sectab_i->VirtualAddress,
2460 sectab_i->SizeOfRawData,
2461 sectab_i->PointerToRawData,
2462 sectab_i->NumberOfRelocations,
2463 sectab_i->PointerToRelocations,
2464 sectab_i->PointerToRawData
2466 reltab = (COFF_reloc*) (
2467 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2470 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2471 /* If the relocation field (a short) has overflowed, the
2472 * real count can be found in the first reloc entry.
2474 * See Section 4.1 (last para) of the PE spec (rev6.0).
2476 COFF_reloc* rel = (COFF_reloc*)
2477 myindex ( sizeof_COFF_reloc, reltab, 0 );
2478 noRelocs = rel->VirtualAddress;
2481 noRelocs = sectab_i->NumberOfRelocations;
2485 for (; j < noRelocs; j++) {
2487 COFF_reloc* rel = (COFF_reloc*)
2488 myindex ( sizeof_COFF_reloc, reltab, j );
2490 " type 0x%-4x vaddr 0x%-8x name `",
2492 rel->VirtualAddress );
2493 sym = (COFF_symbol*)
2494 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
2495 /* Hmm..mysterious looking offset - what's it for? SOF */
2496 printName ( sym->Name, strtab -10 );
2503 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
2504 debugBelch("---START of string table---\n");
2505 for (i = 4; i < *(Int32*)strtab; i++) {
2507 debugBelch("\n"); else
2508 debugBelch("%c", strtab[i] );
2510 debugBelch("--- END of string table---\n");
2515 COFF_symbol* symtab_i;
2516 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2517 symtab_i = (COFF_symbol*)
2518 myindex ( sizeof_COFF_symbol, symtab, i );
2524 printName ( symtab_i->Name, strtab );
2533 (Int32)(symtab_i->SectionNumber),
2534 (UInt32)symtab_i->Type,
2535 (UInt32)symtab_i->StorageClass,
2536 (UInt32)symtab_i->NumberOfAuxSymbols
2538 i += symtab_i->NumberOfAuxSymbols;
2548 ocGetNames_PEi386 ( ObjectCode* oc )
2551 COFF_section* sectab;
2552 COFF_symbol* symtab;
2559 hdr = (COFF_header*)(oc->image);
2560 sectab = (COFF_section*) (
2561 ((UChar*)(oc->image))
2562 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2564 symtab = (COFF_symbol*) (
2565 ((UChar*)(oc->image))
2566 + hdr->PointerToSymbolTable
2568 strtab = ((UChar*)(oc->image))
2569 + hdr->PointerToSymbolTable
2570 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2572 /* Allocate space for any (local, anonymous) .bss sections. */
2574 for (i = 0; i < hdr->NumberOfSections; i++) {
2577 COFF_section* sectab_i
2579 myindex ( sizeof_COFF_section, sectab, i );
2580 if (0 != strcmp((char*)sectab_i->Name, ".bss")) continue;
2581 /* sof 10/05: the PE spec text isn't too clear regarding what
2582 * the SizeOfRawData field is supposed to hold for object
2583 * file sections containing just uninitialized data -- for executables,
2584 * it is supposed to be zero; unclear what it's supposed to be
2585 * for object files. However, VirtualSize is guaranteed to be
2586 * zero for object files, which definitely suggests that SizeOfRawData
2587 * will be non-zero (where else would the size of this .bss section be
2588 * stored?) Looking at the COFF_section info for incoming object files,
2589 * this certainly appears to be the case.
2591 * => I suspect we've been incorrectly handling .bss sections in (relocatable)
2592 * object files up until now. This turned out to bite us with ghc-6.4.1's use
2593 * of gcc-3.4.x, which has started to emit initially-zeroed-out local 'static'
2594 * variable decls into to the .bss section. (The specific function in Q which
2595 * triggered this is libraries/base/cbits/dirUtils.c:__hscore_getFolderPath())
2597 if (sectab_i->VirtualSize == 0 && sectab_i->SizeOfRawData == 0) continue;
2598 /* This is a non-empty .bss section. Allocate zeroed space for
2599 it, and set its PointerToRawData field such that oc->image +
2600 PointerToRawData == addr_of_zeroed_space. */
2601 bss_sz = sectab_i->VirtualSize;
2602 if ( bss_sz < sectab_i->SizeOfRawData) { bss_sz = sectab_i->SizeOfRawData; }
2603 zspace = stgCallocBytes(1, bss_sz, "ocGetNames_PEi386(anonymous bss)");
2604 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
2605 addProddableBlock(oc, zspace, bss_sz);
2606 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
2609 /* Copy section information into the ObjectCode. */
2611 for (i = 0; i < hdr->NumberOfSections; i++) {
2617 = SECTIONKIND_OTHER;
2618 COFF_section* sectab_i
2620 myindex ( sizeof_COFF_section, sectab, i );
2621 IF_DEBUG(linker, debugBelch("section name = %s\n", sectab_i->Name ));
2624 /* I'm sure this is the Right Way to do it. However, the
2625 alternative of testing the sectab_i->Name field seems to
2626 work ok with Cygwin.
2628 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
2629 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
2630 kind = SECTIONKIND_CODE_OR_RODATA;
2633 if (0==strcmp(".text",(char*)sectab_i->Name) ||
2634 0==strcmp(".rdata",(char*)sectab_i->Name)||
2635 0==strcmp(".rodata",(char*)sectab_i->Name))
2636 kind = SECTIONKIND_CODE_OR_RODATA;
2637 if (0==strcmp(".data",(char*)sectab_i->Name) ||
2638 0==strcmp(".bss",(char*)sectab_i->Name))
2639 kind = SECTIONKIND_RWDATA;
2641 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
2642 sz = sectab_i->SizeOfRawData;
2643 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
2645 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
2646 end = start + sz - 1;
2648 if (kind == SECTIONKIND_OTHER
2649 /* Ignore sections called which contain stabs debugging
2651 && 0 != strcmp(".stab", (char*)sectab_i->Name)
2652 && 0 != strcmp(".stabstr", (char*)sectab_i->Name)
2653 /* ignore constructor section for now */
2654 && 0 != strcmp(".ctors", (char*)sectab_i->Name)
2655 /* ignore section generated from .ident */
2656 && 0!= strcmp("/4", (char*)sectab_i->Name)
2657 /* ignore unknown section that appeared in gcc 3.4.5(?) */
2658 && 0!= strcmp(".reloc", (char*)sectab_i->Name)
2660 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName);
2664 if (kind != SECTIONKIND_OTHER && end >= start) {
2665 addSection(oc, kind, start, end);
2666 addProddableBlock(oc, start, end - start + 1);
2670 /* Copy exported symbols into the ObjectCode. */
2672 oc->n_symbols = hdr->NumberOfSymbols;
2673 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2674 "ocGetNames_PEi386(oc->symbols)");
2675 /* Call me paranoid; I don't care. */
2676 for (i = 0; i < oc->n_symbols; i++)
2677 oc->symbols[i] = NULL;
2681 COFF_symbol* symtab_i;
2682 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2683 symtab_i = (COFF_symbol*)
2684 myindex ( sizeof_COFF_symbol, symtab, i );
2688 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2689 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2690 /* This symbol is global and defined, viz, exported */
2691 /* for MYIMAGE_SYMCLASS_EXTERNAL
2692 && !MYIMAGE_SYM_UNDEFINED,
2693 the address of the symbol is:
2694 address of relevant section + offset in section
2696 COFF_section* sectabent
2697 = (COFF_section*) myindex ( sizeof_COFF_section,
2699 symtab_i->SectionNumber-1 );
2700 addr = ((UChar*)(oc->image))
2701 + (sectabent->PointerToRawData
2705 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2706 && symtab_i->Value > 0) {
2707 /* This symbol isn't in any section at all, ie, global bss.
2708 Allocate zeroed space for it. */
2709 addr = stgCallocBytes(1, symtab_i->Value,
2710 "ocGetNames_PEi386(non-anonymous bss)");
2711 addSection(oc, SECTIONKIND_RWDATA, addr,
2712 ((UChar*)addr) + symtab_i->Value - 1);
2713 addProddableBlock(oc, addr, symtab_i->Value);
2714 /* debugBelch("BSS section at 0x%x\n", addr); */
2717 if (addr != NULL ) {
2718 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2719 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2720 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2721 ASSERT(i >= 0 && i < oc->n_symbols);
2722 /* cstring_from_COFF_symbol_name always succeeds. */
2723 oc->symbols[i] = (char*)sname;
2724 ghciInsertStrHashTable(oc->fileName, symhash, (char*)sname, addr);
2728 "IGNORING symbol %d\n"
2732 printName ( symtab_i->Name, strtab );
2741 (Int32)(symtab_i->SectionNumber),
2742 (UInt32)symtab_i->Type,
2743 (UInt32)symtab_i->StorageClass,
2744 (UInt32)symtab_i->NumberOfAuxSymbols
2749 i += symtab_i->NumberOfAuxSymbols;
2758 ocResolve_PEi386 ( ObjectCode* oc )
2761 COFF_section* sectab;
2762 COFF_symbol* symtab;
2772 /* ToDo: should be variable-sized? But is at least safe in the
2773 sense of buffer-overrun-proof. */
2775 /* debugBelch("resolving for %s\n", oc->fileName); */
2777 hdr = (COFF_header*)(oc->image);
2778 sectab = (COFF_section*) (
2779 ((UChar*)(oc->image))
2780 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2782 symtab = (COFF_symbol*) (
2783 ((UChar*)(oc->image))
2784 + hdr->PointerToSymbolTable
2786 strtab = ((UChar*)(oc->image))
2787 + hdr->PointerToSymbolTable
2788 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2790 for (i = 0; i < hdr->NumberOfSections; i++) {
2791 COFF_section* sectab_i
2793 myindex ( sizeof_COFF_section, sectab, i );
2796 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2799 /* Ignore sections called which contain stabs debugging
2801 if (0 == strcmp(".stab", (char*)sectab_i->Name)
2802 || 0 == strcmp(".stabstr", (char*)sectab_i->Name)
2803 || 0 == strcmp(".ctors", (char*)sectab_i->Name))
2806 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2807 /* If the relocation field (a short) has overflowed, the
2808 * real count can be found in the first reloc entry.
2810 * See Section 4.1 (last para) of the PE spec (rev6.0).
2812 * Nov2003 update: the GNU linker still doesn't correctly
2813 * handle the generation of relocatable object files with
2814 * overflown relocations. Hence the output to warn of potential
2817 COFF_reloc* rel = (COFF_reloc*)
2818 myindex ( sizeof_COFF_reloc, reltab, 0 );
2819 noRelocs = rel->VirtualAddress;
2821 /* 10/05: we now assume (and check for) a GNU ld that is capable
2822 * of handling object files with (>2^16) of relocs.
2825 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2830 noRelocs = sectab_i->NumberOfRelocations;
2835 for (; j < noRelocs; j++) {
2837 COFF_reloc* reltab_j
2839 myindex ( sizeof_COFF_reloc, reltab, j );
2841 /* the location to patch */
2843 ((UChar*)(oc->image))
2844 + (sectab_i->PointerToRawData
2845 + reltab_j->VirtualAddress
2846 - sectab_i->VirtualAddress )
2848 /* the existing contents of pP */
2850 /* the symbol to connect to */
2851 sym = (COFF_symbol*)
2852 myindex ( sizeof_COFF_symbol,
2853 symtab, reltab_j->SymbolTableIndex );
2856 "reloc sec %2d num %3d: type 0x%-4x "
2857 "vaddr 0x%-8x name `",
2859 (UInt32)reltab_j->Type,
2860 reltab_j->VirtualAddress );
2861 printName ( sym->Name, strtab );
2862 debugBelch("'\n" ));
2864 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2865 COFF_section* section_sym
2866 = findPEi386SectionCalled ( oc, sym->Name );
2868 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2871 S = ((UInt32)(oc->image))
2872 + (section_sym->PointerToRawData
2875 copyName ( sym->Name, strtab, symbol, 1000-1 );
2876 S = (UInt32) lookupSymbol( (char*)symbol );
2877 if ((void*)S != NULL) goto foundit;
2878 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
2882 checkProddableBlock(oc, pP);
2883 switch (reltab_j->Type) {
2884 case MYIMAGE_REL_I386_DIR32:
2887 case MYIMAGE_REL_I386_REL32:
2888 /* Tricky. We have to insert a displacement at
2889 pP which, when added to the PC for the _next_
2890 insn, gives the address of the target (S).
2891 Problem is to know the address of the next insn
2892 when we only know pP. We assume that this
2893 literal field is always the last in the insn,
2894 so that the address of the next insn is pP+4
2895 -- hence the constant 4.
2896 Also I don't know if A should be added, but so
2897 far it has always been zero.
2899 SOF 05/2005: 'A' (old contents of *pP) have been observed
2900 to contain values other than zero (the 'wx' object file
2901 that came with wxhaskell-0.9.4; dunno how it was compiled..).
2902 So, add displacement to old value instead of asserting
2903 A to be zero. Fixes wxhaskell-related crashes, and no other
2904 ill effects have been observed.
2906 Update: the reason why we're seeing these more elaborate
2907 relocations is due to a switch in how the NCG compiles SRTs
2908 and offsets to them from info tables. SRTs live in .(ro)data,
2909 while info tables live in .text, causing GAS to emit REL32/DISP32
2910 relocations with non-zero values. Adding the displacement is
2911 the right thing to do.
2913 *pP = S - ((UInt32)pP) - 4 + A;
2916 debugBelch("%s: unhandled PEi386 relocation type %d",
2917 oc->fileName, reltab_j->Type);
2924 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2928 #endif /* defined(OBJFORMAT_PEi386) */
2931 /* --------------------------------------------------------------------------
2933 * ------------------------------------------------------------------------*/
2935 #if defined(OBJFORMAT_ELF)
2940 #if defined(sparc_HOST_ARCH)
2941 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2942 #elif defined(i386_HOST_ARCH)
2943 # define ELF_TARGET_386 /* Used inside <elf.h> */
2944 #elif defined(x86_64_HOST_ARCH)
2945 # define ELF_TARGET_X64_64
2949 #if !defined(openbsd_HOST_OS)
2952 /* openbsd elf has things in different places, with diff names */
2953 # include <elf_abi.h>
2954 # include <machine/reloc.h>
2955 # define R_386_32 RELOC_32
2956 # define R_386_PC32 RELOC_PC32
2959 /* If elf.h doesn't define it */
2960 # ifndef R_X86_64_PC64
2961 # define R_X86_64_PC64 24
2965 * Define a set of types which can be used for both ELF32 and ELF64
2969 #define ELFCLASS ELFCLASS64
2970 #define Elf_Addr Elf64_Addr
2971 #define Elf_Word Elf64_Word
2972 #define Elf_Sword Elf64_Sword
2973 #define Elf_Ehdr Elf64_Ehdr
2974 #define Elf_Phdr Elf64_Phdr
2975 #define Elf_Shdr Elf64_Shdr
2976 #define Elf_Sym Elf64_Sym
2977 #define Elf_Rel Elf64_Rel
2978 #define Elf_Rela Elf64_Rela
2980 #define ELF_ST_TYPE ELF64_ST_TYPE
2983 #define ELF_ST_BIND ELF64_ST_BIND
2986 #define ELF_R_TYPE ELF64_R_TYPE
2989 #define ELF_R_SYM ELF64_R_SYM
2992 #define ELFCLASS ELFCLASS32
2993 #define Elf_Addr Elf32_Addr
2994 #define Elf_Word Elf32_Word
2995 #define Elf_Sword Elf32_Sword
2996 #define Elf_Ehdr Elf32_Ehdr
2997 #define Elf_Phdr Elf32_Phdr
2998 #define Elf_Shdr Elf32_Shdr
2999 #define Elf_Sym Elf32_Sym
3000 #define Elf_Rel Elf32_Rel
3001 #define Elf_Rela Elf32_Rela
3003 #define ELF_ST_TYPE ELF32_ST_TYPE
3006 #define ELF_ST_BIND ELF32_ST_BIND
3009 #define ELF_R_TYPE ELF32_R_TYPE
3012 #define ELF_R_SYM ELF32_R_SYM
3018 * Functions to allocate entries in dynamic sections. Currently we simply
3019 * preallocate a large number, and we don't check if a entry for the given
3020 * target already exists (a linear search is too slow). Ideally these
3021 * entries would be associated with symbols.
3024 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
3025 #define GOT_SIZE 0x20000
3026 #define FUNCTION_TABLE_SIZE 0x10000
3027 #define PLT_SIZE 0x08000
3030 static Elf_Addr got[GOT_SIZE];
3031 static unsigned int gotIndex;
3032 static Elf_Addr gp_val = (Elf_Addr)got;
3035 allocateGOTEntry(Elf_Addr target)
3039 if (gotIndex >= GOT_SIZE)
3040 barf("Global offset table overflow");
3042 entry = &got[gotIndex++];
3044 return (Elf_Addr)entry;
3048 #ifdef ELF_FUNCTION_DESC
3054 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
3055 static unsigned int functionTableIndex;
3058 allocateFunctionDesc(Elf_Addr target)
3060 FunctionDesc *entry;
3062 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
3063 barf("Function table overflow");
3065 entry = &functionTable[functionTableIndex++];
3067 entry->gp = (Elf_Addr)gp_val;
3068 return (Elf_Addr)entry;
3072 copyFunctionDesc(Elf_Addr target)
3074 FunctionDesc *olddesc = (FunctionDesc *)target;
3075 FunctionDesc *newdesc;
3077 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
3078 newdesc->gp = olddesc->gp;
3079 return (Elf_Addr)newdesc;
3086 unsigned char code[sizeof(plt_code)];
3090 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
3092 PLTEntry *plt = (PLTEntry *)oc->plt;
3095 if (oc->pltIndex >= PLT_SIZE)
3096 barf("Procedure table overflow");
3098 entry = &plt[oc->pltIndex++];
3099 memcpy(entry->code, plt_code, sizeof(entry->code));
3100 PLT_RELOC(entry->code, target);
3101 return (Elf_Addr)entry;
3107 return (PLT_SIZE * sizeof(PLTEntry));
3113 * Generic ELF functions
3117 findElfSection ( void* objImage, Elf_Word sh_type )
3119 char* ehdrC = (char*)objImage;
3120 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3121 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
3122 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3126 for (i = 0; i < ehdr->e_shnum; i++) {
3127 if (shdr[i].sh_type == sh_type
3128 /* Ignore the section header's string table. */
3129 && i != ehdr->e_shstrndx
3130 /* Ignore string tables named .stabstr, as they contain
3132 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3134 ptr = ehdrC + shdr[i].sh_offset;
3142 ocVerifyImage_ELF ( ObjectCode* oc )
3146 int i, j, nent, nstrtab, nsymtabs;
3150 char* ehdrC = (char*)(oc->image);
3151 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3153 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
3154 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
3155 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
3156 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
3157 errorBelch("%s: not an ELF object", oc->fileName);
3161 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
3162 errorBelch("%s: unsupported ELF format", oc->fileName);
3166 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
3167 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
3169 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
3170 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
3172 errorBelch("%s: unknown endiannness", oc->fileName);
3176 if (ehdr->e_type != ET_REL) {
3177 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
3180 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
3182 IF_DEBUG(linker,debugBelch( "Architecture is " ));
3183 switch (ehdr->e_machine) {
3184 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
3185 #ifdef EM_SPARC32PLUS
3186 case EM_SPARC32PLUS:
3188 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
3190 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
3192 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
3194 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
3195 #elif defined(EM_AMD64)
3196 case EM_AMD64: IF_DEBUG(linker,debugBelch( "amd64" )); break;
3198 default: IF_DEBUG(linker,debugBelch( "unknown" ));
3199 errorBelch("%s: unknown architecture (e_machine == %d)"
3200 , oc->fileName, ehdr->e_machine);
3204 IF_DEBUG(linker,debugBelch(
3205 "\nSection header table: start %ld, n_entries %d, ent_size %d\n",
3206 (long)ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
3208 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
3210 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3212 if (ehdr->e_shstrndx == SHN_UNDEF) {
3213 errorBelch("%s: no section header string table", oc->fileName);
3216 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
3218 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3221 for (i = 0; i < ehdr->e_shnum; i++) {
3222 IF_DEBUG(linker,debugBelch("%2d: ", i ));
3223 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
3224 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
3225 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
3226 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
3227 ehdrC + shdr[i].sh_offset,
3228 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
3230 if (shdr[i].sh_type == SHT_REL) {
3231 IF_DEBUG(linker,debugBelch("Rel " ));
3232 } else if (shdr[i].sh_type == SHT_RELA) {
3233 IF_DEBUG(linker,debugBelch("RelA " ));
3235 IF_DEBUG(linker,debugBelch(" "));
3238 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
3242 IF_DEBUG(linker,debugBelch( "\nString tables" ));
3245 for (i = 0; i < ehdr->e_shnum; i++) {
3246 if (shdr[i].sh_type == SHT_STRTAB
3247 /* Ignore the section header's string table. */
3248 && i != ehdr->e_shstrndx
3249 /* Ignore string tables named .stabstr, as they contain
3251 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3253 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
3254 strtab = ehdrC + shdr[i].sh_offset;
3259 errorBelch("%s: no string tables, or too many", oc->fileName);
3264 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
3265 for (i = 0; i < ehdr->e_shnum; i++) {
3266 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3267 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
3269 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3270 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3271 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%ld rem)\n",
3273 (long)shdr[i].sh_size % sizeof(Elf_Sym)
3275 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
3276 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
3279 for (j = 0; j < nent; j++) {
3280 IF_DEBUG(linker,debugBelch(" %2d ", j ));
3281 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
3282 (int)stab[j].st_shndx,
3283 (int)stab[j].st_size,
3284 (char*)stab[j].st_value ));
3286 IF_DEBUG(linker,debugBelch("type=" ));
3287 switch (ELF_ST_TYPE(stab[j].st_info)) {
3288 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
3289 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
3290 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
3291 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
3292 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
3293 default: IF_DEBUG(linker,debugBelch("? " )); break;
3295 IF_DEBUG(linker,debugBelch(" " ));
3297 IF_DEBUG(linker,debugBelch("bind=" ));
3298 switch (ELF_ST_BIND(stab[j].st_info)) {
3299 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
3300 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
3301 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
3302 default: IF_DEBUG(linker,debugBelch("? " )); break;
3304 IF_DEBUG(linker,debugBelch(" " ));
3306 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
3310 if (nsymtabs == 0) {
3311 errorBelch("%s: didn't find any symbol tables", oc->fileName);
3318 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
3322 if (hdr->sh_type == SHT_PROGBITS
3323 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
3324 /* .text-style section */
3325 return SECTIONKIND_CODE_OR_RODATA;
3328 if (hdr->sh_type == SHT_PROGBITS
3329 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3330 /* .data-style section */
3331 return SECTIONKIND_RWDATA;
3334 if (hdr->sh_type == SHT_PROGBITS
3335 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
3336 /* .rodata-style section */
3337 return SECTIONKIND_CODE_OR_RODATA;
3340 if (hdr->sh_type == SHT_NOBITS
3341 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3342 /* .bss-style section */
3344 return SECTIONKIND_RWDATA;
3347 return SECTIONKIND_OTHER;
3352 ocGetNames_ELF ( ObjectCode* oc )
3357 char* ehdrC = (char*)(oc->image);
3358 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3359 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
3360 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3362 ASSERT(symhash != NULL);
3365 errorBelch("%s: no strtab", oc->fileName);
3370 for (i = 0; i < ehdr->e_shnum; i++) {
3371 /* Figure out what kind of section it is. Logic derived from
3372 Figure 1.14 ("Special Sections") of the ELF document
3373 ("Portable Formats Specification, Version 1.1"). */
3375 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
3377 if (is_bss && shdr[i].sh_size > 0) {
3378 /* This is a non-empty .bss section. Allocate zeroed space for
3379 it, and set its .sh_offset field such that
3380 ehdrC + .sh_offset == addr_of_zeroed_space. */
3381 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
3382 "ocGetNames_ELF(BSS)");
3383 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
3385 debugBelch("BSS section at 0x%x, size %d\n",
3386 zspace, shdr[i].sh_size);
3390 /* fill in the section info */
3391 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
3392 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
3393 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
3394 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
3397 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3399 /* copy stuff into this module's object symbol table */
3400 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3401 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3403 oc->n_symbols = nent;
3404 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3405 "ocGetNames_ELF(oc->symbols)");
3407 for (j = 0; j < nent; j++) {
3409 char isLocal = FALSE; /* avoids uninit-var warning */
3411 char* nm = strtab + stab[j].st_name;
3412 int secno = stab[j].st_shndx;
3414 /* Figure out if we want to add it; if so, set ad to its
3415 address. Otherwise leave ad == NULL. */
3417 if (secno == SHN_COMMON) {
3419 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
3421 debugBelch("COMMON symbol, size %d name %s\n",
3422 stab[j].st_size, nm);
3424 /* Pointless to do addProddableBlock() for this area,
3425 since the linker should never poke around in it. */
3428 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
3429 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
3431 /* and not an undefined symbol */
3432 && stab[j].st_shndx != SHN_UNDEF
3433 /* and not in a "special section" */
3434 && stab[j].st_shndx < SHN_LORESERVE
3436 /* and it's a not a section or string table or anything silly */
3437 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
3438 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
3439 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
3442 /* Section 0 is the undefined section, hence > and not >=. */
3443 ASSERT(secno > 0 && secno < ehdr->e_shnum);
3445 if (shdr[secno].sh_type == SHT_NOBITS) {
3446 debugBelch(" BSS symbol, size %d off %d name %s\n",
3447 stab[j].st_size, stab[j].st_value, nm);
3450 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
3451 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
3454 #ifdef ELF_FUNCTION_DESC
3455 /* dlsym() and the initialisation table both give us function
3456 * descriptors, so to be consistent we store function descriptors
3457 * in the symbol table */
3458 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
3459 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
3461 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s\n",
3462 ad, oc->fileName, nm ));
3467 /* And the decision is ... */
3471 oc->symbols[j] = nm;
3474 /* Ignore entirely. */
3476 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
3480 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
3481 strtab + stab[j].st_name ));
3484 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
3485 (int)ELF_ST_BIND(stab[j].st_info),
3486 (int)ELF_ST_TYPE(stab[j].st_info),
3487 (int)stab[j].st_shndx,
3488 strtab + stab[j].st_name
3491 oc->symbols[j] = NULL;
3500 /* Do ELF relocations which lack an explicit addend. All x86-linux
3501 relocations appear to be of this form. */
3503 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
3504 Elf_Shdr* shdr, int shnum,
3505 Elf_Sym* stab, char* strtab )
3510 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
3511 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
3512 int target_shndx = shdr[shnum].sh_info;
3513 int symtab_shndx = shdr[shnum].sh_link;
3515 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3516 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
3517 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3518 target_shndx, symtab_shndx ));
3520 /* Skip sections that we're not interested in. */
3523 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
3524 if (kind == SECTIONKIND_OTHER) {
3525 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
3530 for (j = 0; j < nent; j++) {
3531 Elf_Addr offset = rtab[j].r_offset;
3532 Elf_Addr info = rtab[j].r_info;
3534 Elf_Addr P = ((Elf_Addr)targ) + offset;
3535 Elf_Word* pP = (Elf_Word*)P;
3540 StgStablePtr stablePtr;
3543 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
3544 j, (void*)offset, (void*)info ));
3546 IF_DEBUG(linker,debugBelch( " ZERO" ));
3549 Elf_Sym sym = stab[ELF_R_SYM(info)];
3550 /* First see if it is a local symbol. */
3551 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3552 /* Yes, so we can get the address directly from the ELF symbol
3554 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3556 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3557 + stab[ELF_R_SYM(info)].st_value);
3560 symbol = strtab + sym.st_name;
3561 stablePtr = (StgStablePtr)lookupHashTable(stablehash, (StgWord)symbol);
3562 if (NULL == stablePtr) {
3563 /* No, so look up the name in our global table. */
3564 S_tmp = lookupSymbol( symbol );
3565 S = (Elf_Addr)S_tmp;
3567 stableVal = deRefStablePtr( stablePtr );
3569 S = (Elf_Addr)S_tmp;
3573 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3576 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
3579 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
3580 (void*)P, (void*)S, (void*)A ));
3581 checkProddableBlock ( oc, pP );
3585 switch (ELF_R_TYPE(info)) {
3586 # ifdef i386_HOST_ARCH
3587 case R_386_32: *pP = value; break;
3588 case R_386_PC32: *pP = value - P; break;
3591 errorBelch("%s: unhandled ELF relocation(Rel) type %lu\n",
3592 oc->fileName, (lnat)ELF_R_TYPE(info));
3600 /* Do ELF relocations for which explicit addends are supplied.
3601 sparc-solaris relocations appear to be of this form. */
3603 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
3604 Elf_Shdr* shdr, int shnum,
3605 Elf_Sym* stab, char* strtab )
3608 char *symbol = NULL;
3610 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
3611 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
3612 int target_shndx = shdr[shnum].sh_info;
3613 int symtab_shndx = shdr[shnum].sh_link;
3615 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3616 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
3617 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3618 target_shndx, symtab_shndx ));
3620 for (j = 0; j < nent; j++) {
3621 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3622 /* This #ifdef only serves to avoid unused-var warnings. */
3623 Elf_Addr offset = rtab[j].r_offset;
3624 Elf_Addr P = targ + offset;
3626 Elf_Addr info = rtab[j].r_info;
3627 Elf_Addr A = rtab[j].r_addend;
3631 # if defined(sparc_HOST_ARCH)
3632 Elf_Word* pP = (Elf_Word*)P;
3634 # elif defined(powerpc_HOST_ARCH)
3638 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
3639 j, (void*)offset, (void*)info,
3642 IF_DEBUG(linker,debugBelch( " ZERO" ));
3645 Elf_Sym sym = stab[ELF_R_SYM(info)];
3646 /* First see if it is a local symbol. */
3647 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3648 /* Yes, so we can get the address directly from the ELF symbol
3650 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3652 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3653 + stab[ELF_R_SYM(info)].st_value);
3654 #ifdef ELF_FUNCTION_DESC
3655 /* Make a function descriptor for this function */
3656 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
3657 S = allocateFunctionDesc(S + A);
3662 /* No, so look up the name in our global table. */
3663 symbol = strtab + sym.st_name;
3664 S_tmp = lookupSymbol( symbol );
3665 S = (Elf_Addr)S_tmp;
3667 #ifdef ELF_FUNCTION_DESC
3668 /* If a function, already a function descriptor - we would
3669 have to copy it to add an offset. */
3670 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3671 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3675 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3678 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3681 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3682 (void*)P, (void*)S, (void*)A ));
3683 /* checkProddableBlock ( oc, (void*)P ); */
3687 switch (ELF_R_TYPE(info)) {
3688 # if defined(sparc_HOST_ARCH)
3689 case R_SPARC_WDISP30:
3690 w1 = *pP & 0xC0000000;
3691 w2 = (Elf_Word)((value - P) >> 2);
3692 ASSERT((w2 & 0xC0000000) == 0);
3697 w1 = *pP & 0xFFC00000;
3698 w2 = (Elf_Word)(value >> 10);
3699 ASSERT((w2 & 0xFFC00000) == 0);
3705 w2 = (Elf_Word)(value & 0x3FF);
3706 ASSERT((w2 & ~0x3FF) == 0);
3711 /* According to the Sun documentation:
3713 This relocation type resembles R_SPARC_32, except it refers to an
3714 unaligned word. That is, the word to be relocated must be treated
3715 as four separate bytes with arbitrary alignment, not as a word
3716 aligned according to the architecture requirements.
3719 w2 = (Elf_Word)value;
3721 // SPARC doesn't do misaligned writes of 32 bit words,
3722 // so we have to do this one byte-at-a-time.
3723 char *pPc = (char*)pP;
3724 pPc[0] = (char) ((Elf_Word)(w2 & 0xff000000) >> 24);
3725 pPc[1] = (char) ((Elf_Word)(w2 & 0x00ff0000) >> 16);
3726 pPc[2] = (char) ((Elf_Word)(w2 & 0x0000ff00) >> 8);
3727 pPc[3] = (char) ((Elf_Word)(w2 & 0x000000ff));
3731 w2 = (Elf_Word)value;
3734 # elif defined(powerpc_HOST_ARCH)
3735 case R_PPC_ADDR16_LO:
3736 *(Elf32_Half*) P = value;
3739 case R_PPC_ADDR16_HI:
3740 *(Elf32_Half*) P = value >> 16;
3743 case R_PPC_ADDR16_HA:
3744 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3748 *(Elf32_Word *) P = value;
3752 *(Elf32_Word *) P = value - P;
3758 if( delta << 6 >> 6 != delta )
3760 value = (Elf_Addr) (&makeSymbolExtra( oc, ELF_R_SYM(info), value )
3764 if( value == 0 || delta << 6 >> 6 != delta )
3766 barf( "Unable to make SymbolExtra for #%d",
3772 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3773 | (delta & 0x3fffffc);
3777 #if x86_64_HOST_ARCH
3779 *(Elf64_Xword *)P = value;
3784 #if defined(ALWAYS_PIC)
3785 barf("R_X86_64_PC32 relocation, but ALWAYS_PIC.");
3787 StgInt64 off = value - P;
3788 if (off >= 0x7fffffffL || off < -0x80000000L) {
3789 #if X86_64_ELF_NONPIC_HACK
3790 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3792 off = pltAddress + A - P;
3794 barf("R_X86_64_PC32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3795 symbol, off, oc->fileName );
3798 *(Elf64_Word *)P = (Elf64_Word)off;
3805 StgInt64 off = value - P;
3806 *(Elf64_Word *)P = (Elf64_Word)off;
3811 #if defined(ALWAYS_PIC)
3812 barf("R_X86_64_32 relocation, but ALWAYS_PIC.");
3814 if (value >= 0x7fffffffL) {
3815 #if X86_64_ELF_NONPIC_HACK
3816 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3818 value = pltAddress + A;
3820 barf("R_X86_64_32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3821 symbol, value, oc->fileName );
3824 *(Elf64_Word *)P = (Elf64_Word)value;
3829 #if defined(ALWAYS_PIC)
3830 barf("R_X86_64_32S relocation, but ALWAYS_PIC.");
3832 if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) {
3833 #if X86_64_ELF_NONPIC_HACK
3834 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3836 value = pltAddress + A;
3838 barf("R_X86_64_32S relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3839 symbol, value, oc->fileName );
3842 *(Elf64_Sword *)P = (Elf64_Sword)value;
3846 case R_X86_64_GOTPCREL:
3848 StgInt64 gotAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)->addr;
3849 StgInt64 off = gotAddress + A - P;
3850 *(Elf64_Word *)P = (Elf64_Word)off;
3854 case R_X86_64_PLT32:
3856 #if defined(ALWAYS_PIC)
3857 barf("R_X86_64_PLT32 relocation, but ALWAYS_PIC.");
3859 StgInt64 off = value - P;
3860 if (off >= 0x7fffffffL || off < -0x80000000L) {
3861 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3863 off = pltAddress + A - P;
3865 *(Elf64_Word *)P = (Elf64_Word)off;
3872 errorBelch("%s: unhandled ELF relocation(RelA) type %lu\n",
3873 oc->fileName, (lnat)ELF_R_TYPE(info));
3882 ocResolve_ELF ( ObjectCode* oc )
3886 Elf_Sym* stab = NULL;
3887 char* ehdrC = (char*)(oc->image);
3888 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3889 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3891 /* first find "the" symbol table */
3892 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3894 /* also go find the string table */
3895 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3897 if (stab == NULL || strtab == NULL) {
3898 errorBelch("%s: can't find string or symbol table", oc->fileName);
3902 /* Process the relocation sections. */
3903 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3904 if (shdr[shnum].sh_type == SHT_REL) {
3905 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3906 shnum, stab, strtab );
3910 if (shdr[shnum].sh_type == SHT_RELA) {
3911 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3912 shnum, stab, strtab );
3917 #if defined(powerpc_HOST_ARCH)
3918 ocFlushInstructionCache( oc );
3925 * PowerPC & X86_64 ELF specifics
3928 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3930 static int ocAllocateSymbolExtras_ELF( ObjectCode *oc )
3936 ehdr = (Elf_Ehdr *) oc->image;
3937 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3939 for( i = 0; i < ehdr->e_shnum; i++ )
3940 if( shdr[i].sh_type == SHT_SYMTAB )
3943 if( i == ehdr->e_shnum )
3945 errorBelch( "This ELF file contains no symtab" );
3949 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3951 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3952 (int) shdr[i].sh_entsize, (int) sizeof( Elf_Sym ) );
3957 return ocAllocateSymbolExtras( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3960 #endif /* powerpc */
3964 /* --------------------------------------------------------------------------
3966 * ------------------------------------------------------------------------*/
3968 #if defined(OBJFORMAT_MACHO)
3971 Support for MachO linking on Darwin/MacOS X
3972 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3974 I hereby formally apologize for the hackish nature of this code.
3975 Things that need to be done:
3976 *) implement ocVerifyImage_MachO
3977 *) add still more sanity checks.
3980 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
3981 #define mach_header mach_header_64
3982 #define segment_command segment_command_64
3983 #define section section_64
3984 #define nlist nlist_64
3987 #ifdef powerpc_HOST_ARCH
3988 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3990 struct mach_header *header = (struct mach_header *) oc->image;
3991 struct load_command *lc = (struct load_command *) (header + 1);
3994 for( i = 0; i < header->ncmds; i++ )
3996 if( lc->cmd == LC_SYMTAB )
3998 // Find out the first and last undefined external
3999 // symbol, so we don't have to allocate too many
4001 struct symtab_command *symLC = (struct symtab_command *) lc;
4002 unsigned min = symLC->nsyms, max = 0;
4003 struct nlist *nlist =
4004 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
4006 for(i=0;i<symLC->nsyms;i++)
4008 if(nlist[i].n_type & N_STAB)
4010 else if(nlist[i].n_type & N_EXT)
4012 if((nlist[i].n_type & N_TYPE) == N_UNDF
4013 && (nlist[i].n_value == 0))
4023 return ocAllocateSymbolExtras(oc, max - min + 1, min);
4028 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
4030 return ocAllocateSymbolExtras(oc,0,0);
4033 #ifdef x86_64_HOST_ARCH
4034 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
4036 struct mach_header *header = (struct mach_header *) oc->image;
4037 struct load_command *lc = (struct load_command *) (header + 1);
4040 for( i = 0; i < header->ncmds; i++ )
4042 if( lc->cmd == LC_SYMTAB )
4044 // Just allocate one entry for every symbol
4045 struct symtab_command *symLC = (struct symtab_command *) lc;
4047 return ocAllocateSymbolExtras(oc, symLC->nsyms, 0);
4050 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
4052 return ocAllocateSymbolExtras(oc,0,0);
4056 static int ocVerifyImage_MachO(ObjectCode* oc)
4058 char *image = (char*) oc->image;
4059 struct mach_header *header = (struct mach_header*) image;
4061 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
4062 if(header->magic != MH_MAGIC_64) {
4063 errorBelch("%s: Bad magic. Expected: %08x, got: %08x.\n",
4064 oc->fileName, MH_MAGIC_64, header->magic);
4068 if(header->magic != MH_MAGIC) {
4069 errorBelch("%s: Bad magic. Expected: %08x, got: %08x.\n",
4070 oc->fileName, MH_MAGIC, header->magic);
4074 // FIXME: do some more verifying here
4078 static int resolveImports(
4081 struct symtab_command *symLC,
4082 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
4083 unsigned long *indirectSyms,
4084 struct nlist *nlist)
4087 size_t itemSize = 4;
4090 int isJumpTable = 0;
4091 if(!strcmp(sect->sectname,"__jump_table"))
4095 ASSERT(sect->reserved2 == itemSize);
4099 for(i=0; i*itemSize < sect->size;i++)
4101 // according to otool, reserved1 contains the first index into the indirect symbol table
4102 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
4103 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4106 if((symbol->n_type & N_TYPE) == N_UNDF
4107 && (symbol->n_type & N_EXT) && (symbol->n_value != 0))
4108 addr = (void*) (symbol->n_value);
4110 addr = lookupSymbol(nm);
4113 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
4121 checkProddableBlock(oc,image + sect->offset + i*itemSize);
4122 *(image + sect->offset + i*itemSize) = 0xe9; // jmp
4123 *(unsigned*)(image + sect->offset + i*itemSize + 1)
4124 = (char*)addr - (image + sect->offset + i*itemSize + 5);
4129 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
4130 ((void**)(image + sect->offset))[i] = addr;
4137 static unsigned long relocateAddress(
4140 struct section* sections,
4141 unsigned long address)
4144 for(i = 0; i < nSections; i++)
4146 if(sections[i].addr <= address
4147 && address < sections[i].addr + sections[i].size)
4149 return (unsigned long)oc->image
4150 + sections[i].offset + address - sections[i].addr;
4153 barf("Invalid Mach-O file:"
4154 "Address out of bounds while relocating object file");
4158 static int relocateSection(
4161 struct symtab_command *symLC, struct nlist *nlist,
4162 int nSections, struct section* sections, struct section *sect)
4164 struct relocation_info *relocs;
4167 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
4169 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
4171 else if(!strcmp(sect->sectname,"__la_sym_ptr2"))
4173 else if(!strcmp(sect->sectname,"__la_sym_ptr3"))
4177 relocs = (struct relocation_info*) (image + sect->reloff);
4181 #ifdef x86_64_HOST_ARCH
4182 struct relocation_info *reloc = &relocs[i];
4184 char *thingPtr = image + sect->offset + reloc->r_address;
4186 /* We shouldn't need to initialise this, but gcc on OS X 64 bit
4187 complains that it may be used uninitialized if we don't */
4190 int type = reloc->r_type;
4192 checkProddableBlock(oc,thingPtr);
4193 switch(reloc->r_length)
4196 thing = *(uint8_t*)thingPtr;
4197 baseValue = (uint64_t)thingPtr + 1;
4200 thing = *(uint16_t*)thingPtr;
4201 baseValue = (uint64_t)thingPtr + 2;
4204 thing = *(uint32_t*)thingPtr;
4205 baseValue = (uint64_t)thingPtr + 4;
4208 thing = *(uint64_t*)thingPtr;
4209 baseValue = (uint64_t)thingPtr + 8;
4212 barf("Unknown size.");
4215 if(type == X86_64_RELOC_GOT
4216 || type == X86_64_RELOC_GOT_LOAD)
4218 ASSERT(reloc->r_extern);
4219 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)->addr;
4221 type = X86_64_RELOC_SIGNED;
4223 else if(reloc->r_extern)
4225 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4226 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4227 if(symbol->n_value == 0)
4228 value = (uint64_t) lookupSymbol(nm);
4230 value = relocateAddress(oc, nSections, sections,
4235 value = sections[reloc->r_symbolnum-1].offset
4236 - sections[reloc->r_symbolnum-1].addr
4240 if(type == X86_64_RELOC_BRANCH)
4242 if((int32_t)(value - baseValue) != (int64_t)(value - baseValue))
4244 ASSERT(reloc->r_extern);
4245 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)
4248 ASSERT((int32_t)(value - baseValue) == (int64_t)(value - baseValue));
4249 type = X86_64_RELOC_SIGNED;
4254 case X86_64_RELOC_UNSIGNED:
4255 ASSERT(!reloc->r_pcrel);
4258 case X86_64_RELOC_SIGNED:
4259 case X86_64_RELOC_SIGNED_1:
4260 case X86_64_RELOC_SIGNED_2:
4261 case X86_64_RELOC_SIGNED_4:
4262 ASSERT(reloc->r_pcrel);
4263 thing += value - baseValue;
4265 case X86_64_RELOC_SUBTRACTOR:
4266 ASSERT(!reloc->r_pcrel);
4270 barf("unkown relocation");
4273 switch(reloc->r_length)
4276 *(uint8_t*)thingPtr = thing;
4279 *(uint16_t*)thingPtr = thing;
4282 *(uint32_t*)thingPtr = thing;
4285 *(uint64_t*)thingPtr = thing;
4289 if(relocs[i].r_address & R_SCATTERED)
4291 struct scattered_relocation_info *scat =
4292 (struct scattered_relocation_info*) &relocs[i];
4296 if(scat->r_length == 2)
4298 unsigned long word = 0;
4299 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
4300 checkProddableBlock(oc,wordPtr);
4302 // Note on relocation types:
4303 // i386 uses the GENERIC_RELOC_* types,
4304 // while ppc uses special PPC_RELOC_* types.
4305 // *_RELOC_VANILLA and *_RELOC_PAIR have the same value
4306 // in both cases, all others are different.
4307 // Therefore, we use GENERIC_RELOC_VANILLA
4308 // and GENERIC_RELOC_PAIR instead of the PPC variants,
4309 // and use #ifdefs for the other types.
4311 // Step 1: Figure out what the relocated value should be
4312 if(scat->r_type == GENERIC_RELOC_VANILLA)
4314 word = *wordPtr + (unsigned long) relocateAddress(
4321 #ifdef powerpc_HOST_ARCH
4322 else if(scat->r_type == PPC_RELOC_SECTDIFF
4323 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
4324 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
4325 || scat->r_type == PPC_RELOC_HA16_SECTDIFF
4326 || scat->r_type == PPC_RELOC_LOCAL_SECTDIFF)
4328 else if(scat->r_type == GENERIC_RELOC_SECTDIFF
4329 || scat->r_type == GENERIC_RELOC_LOCAL_SECTDIFF)
4332 struct scattered_relocation_info *pair =
4333 (struct scattered_relocation_info*) &relocs[i+1];
4335 if(!pair->r_scattered || pair->r_type != GENERIC_RELOC_PAIR)
4336 barf("Invalid Mach-O file: "
4337 "RELOC_*_SECTDIFF not followed by RELOC_PAIR");
4339 word = (unsigned long)
4340 (relocateAddress(oc, nSections, sections, scat->r_value)
4341 - relocateAddress(oc, nSections, sections, pair->r_value));
4344 #ifdef powerpc_HOST_ARCH
4345 else if(scat->r_type == PPC_RELOC_HI16
4346 || scat->r_type == PPC_RELOC_LO16
4347 || scat->r_type == PPC_RELOC_HA16
4348 || scat->r_type == PPC_RELOC_LO14)
4349 { // these are generated by label+offset things
4350 struct relocation_info *pair = &relocs[i+1];
4351 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
4352 barf("Invalid Mach-O file: "
4353 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
4355 if(scat->r_type == PPC_RELOC_LO16)
4357 word = ((unsigned short*) wordPtr)[1];
4358 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4360 else if(scat->r_type == PPC_RELOC_LO14)
4362 barf("Unsupported Relocation: PPC_RELOC_LO14");
4363 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
4364 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4366 else if(scat->r_type == PPC_RELOC_HI16)
4368 word = ((unsigned short*) wordPtr)[1] << 16;
4369 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4371 else if(scat->r_type == PPC_RELOC_HA16)
4373 word = ((unsigned short*) wordPtr)[1] << 16;
4374 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4378 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
4386 barf ("Don't know how to handle this Mach-O "
4387 "scattered relocation entry: "
4388 "object file %s; entry type %ld; "
4390 oc->fileName, scat->r_type, scat->r_address);
4394 #ifdef powerpc_HOST_ARCH
4395 if(scat->r_type == GENERIC_RELOC_VANILLA
4396 || scat->r_type == PPC_RELOC_SECTDIFF)
4398 if(scat->r_type == GENERIC_RELOC_VANILLA
4399 || scat->r_type == GENERIC_RELOC_SECTDIFF
4400 || scat->r_type == GENERIC_RELOC_LOCAL_SECTDIFF)
4405 #ifdef powerpc_HOST_ARCH
4406 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
4408 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4410 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
4412 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4414 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
4416 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4417 + ((word & (1<<15)) ? 1 : 0);
4423 barf("Can't handle Mach-O scattered relocation entry "
4424 "with this r_length tag: "
4425 "object file %s; entry type %ld; "
4426 "r_length tag %ld; address %#lx\n",
4427 oc->fileName, scat->r_type, scat->r_length,
4432 else /* scat->r_pcrel */
4434 barf("Don't know how to handle *PC-relative* Mach-O "
4435 "scattered relocation entry: "
4436 "object file %s; entry type %ld; address %#lx\n",
4437 oc->fileName, scat->r_type, scat->r_address);
4442 else /* !(relocs[i].r_address & R_SCATTERED) */
4444 struct relocation_info *reloc = &relocs[i];
4445 if(reloc->r_pcrel && !reloc->r_extern)
4448 if(reloc->r_length == 2)
4450 unsigned long word = 0;
4451 #ifdef powerpc_HOST_ARCH
4452 unsigned long jumpIsland = 0;
4453 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
4454 // to avoid warning and to catch
4458 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
4459 checkProddableBlock(oc,wordPtr);
4461 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4465 #ifdef powerpc_HOST_ARCH
4466 else if(reloc->r_type == PPC_RELOC_LO16)
4468 word = ((unsigned short*) wordPtr)[1];
4469 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4471 else if(reloc->r_type == PPC_RELOC_HI16)
4473 word = ((unsigned short*) wordPtr)[1] << 16;
4474 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4476 else if(reloc->r_type == PPC_RELOC_HA16)
4478 word = ((unsigned short*) wordPtr)[1] << 16;
4479 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4481 else if(reloc->r_type == PPC_RELOC_BR24)
4484 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
4489 barf("Can't handle this Mach-O relocation entry "
4491 "object file %s; entry type %ld; address %#lx\n",
4492 oc->fileName, reloc->r_type, reloc->r_address);
4496 if(!reloc->r_extern)
4499 sections[reloc->r_symbolnum-1].offset
4500 - sections[reloc->r_symbolnum-1].addr
4507 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4508 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4509 void *symbolAddress = lookupSymbol(nm);
4512 errorBelch("\nunknown symbol `%s'", nm);
4518 #ifdef powerpc_HOST_ARCH
4519 // In the .o file, this should be a relative jump to NULL
4520 // and we'll change it to a relative jump to the symbol
4521 ASSERT(word + reloc->r_address == 0);
4522 jumpIsland = (unsigned long)
4523 &makeSymbolExtra(oc,
4525 (unsigned long) symbolAddress)
4529 offsetToJumpIsland = word + jumpIsland
4530 - (((long)image) + sect->offset - sect->addr);
4533 word += (unsigned long) symbolAddress
4534 - (((long)image) + sect->offset - sect->addr);
4538 word += (unsigned long) symbolAddress;
4542 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4547 #ifdef powerpc_HOST_ARCH
4548 else if(reloc->r_type == PPC_RELOC_LO16)
4550 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4553 else if(reloc->r_type == PPC_RELOC_HI16)
4555 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4558 else if(reloc->r_type == PPC_RELOC_HA16)
4560 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4561 + ((word & (1<<15)) ? 1 : 0);
4564 else if(reloc->r_type == PPC_RELOC_BR24)
4566 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4568 // The branch offset is too large.
4569 // Therefore, we try to use a jump island.
4572 barf("unconditional relative branch out of range: "
4573 "no jump island available");
4576 word = offsetToJumpIsland;
4577 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4578 barf("unconditional relative branch out of range: "
4579 "jump island out of range");
4581 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
4588 barf("Can't handle Mach-O relocation entry (not scattered) "
4589 "with this r_length tag: "
4590 "object file %s; entry type %ld; "
4591 "r_length tag %ld; address %#lx\n",
4592 oc->fileName, reloc->r_type, reloc->r_length,
4602 static int ocGetNames_MachO(ObjectCode* oc)
4604 char *image = (char*) oc->image;
4605 struct mach_header *header = (struct mach_header*) image;
4606 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4607 unsigned i,curSymbol = 0;
4608 struct segment_command *segLC = NULL;
4609 struct section *sections;
4610 struct symtab_command *symLC = NULL;
4611 struct nlist *nlist;
4612 unsigned long commonSize = 0;
4613 char *commonStorage = NULL;
4614 unsigned long commonCounter;
4616 for(i=0;i<header->ncmds;i++)
4618 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4619 segLC = (struct segment_command*) lc;
4620 else if(lc->cmd == LC_SYMTAB)
4621 symLC = (struct symtab_command*) lc;
4622 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4625 sections = (struct section*) (segLC+1);
4626 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4630 barf("ocGetNames_MachO: no segment load command");
4632 for(i=0;i<segLC->nsects;i++)
4634 if(sections[i].size == 0)
4637 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
4639 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
4640 "ocGetNames_MachO(common symbols)");
4641 sections[i].offset = zeroFillArea - image;
4644 if(!strcmp(sections[i].sectname,"__text"))
4645 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
4646 (void*) (image + sections[i].offset),
4647 (void*) (image + sections[i].offset + sections[i].size));
4648 else if(!strcmp(sections[i].sectname,"__const"))
4649 addSection(oc, SECTIONKIND_RWDATA,
4650 (void*) (image + sections[i].offset),
4651 (void*) (image + sections[i].offset + sections[i].size));
4652 else if(!strcmp(sections[i].sectname,"__data"))
4653 addSection(oc, SECTIONKIND_RWDATA,
4654 (void*) (image + sections[i].offset),
4655 (void*) (image + sections[i].offset + sections[i].size));
4656 else if(!strcmp(sections[i].sectname,"__bss")
4657 || !strcmp(sections[i].sectname,"__common"))
4658 addSection(oc, SECTIONKIND_RWDATA,
4659 (void*) (image + sections[i].offset),
4660 (void*) (image + sections[i].offset + sections[i].size));
4662 addProddableBlock(oc, (void*) (image + sections[i].offset),
4666 // count external symbols defined here
4670 for(i=0;i<symLC->nsyms;i++)
4672 if(nlist[i].n_type & N_STAB)
4674 else if(nlist[i].n_type & N_EXT)
4676 if((nlist[i].n_type & N_TYPE) == N_UNDF
4677 && (nlist[i].n_value != 0))
4679 commonSize += nlist[i].n_value;
4682 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4687 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
4688 "ocGetNames_MachO(oc->symbols)");
4692 for(i=0;i<symLC->nsyms;i++)
4694 if(nlist[i].n_type & N_STAB)
4696 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4698 if(nlist[i].n_type & N_EXT)
4700 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4701 if((nlist[i].n_desc & N_WEAK_DEF) && lookupSymbol(nm))
4702 ; // weak definition, and we already have a definition
4705 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4707 + sections[nlist[i].n_sect-1].offset
4708 - sections[nlist[i].n_sect-1].addr
4709 + nlist[i].n_value);
4710 oc->symbols[curSymbol++] = nm;
4717 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
4718 commonCounter = (unsigned long)commonStorage;
4721 for(i=0;i<symLC->nsyms;i++)
4723 if((nlist[i].n_type & N_TYPE) == N_UNDF
4724 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
4726 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4727 unsigned long sz = nlist[i].n_value;
4729 nlist[i].n_value = commonCounter;
4731 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4732 (void*)commonCounter);
4733 oc->symbols[curSymbol++] = nm;
4735 commonCounter += sz;
4742 static int ocResolve_MachO(ObjectCode* oc)
4744 char *image = (char*) oc->image;
4745 struct mach_header *header = (struct mach_header*) image;
4746 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4748 struct segment_command *segLC = NULL;
4749 struct section *sections;
4750 struct symtab_command *symLC = NULL;
4751 struct dysymtab_command *dsymLC = NULL;
4752 struct nlist *nlist;
4754 for(i=0;i<header->ncmds;i++)
4756 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4757 segLC = (struct segment_command*) lc;
4758 else if(lc->cmd == LC_SYMTAB)
4759 symLC = (struct symtab_command*) lc;
4760 else if(lc->cmd == LC_DYSYMTAB)
4761 dsymLC = (struct dysymtab_command*) lc;
4762 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4765 sections = (struct section*) (segLC+1);
4766 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4771 unsigned long *indirectSyms
4772 = (unsigned long*) (image + dsymLC->indirectsymoff);
4774 for(i=0;i<segLC->nsects;i++)
4776 if( !strcmp(sections[i].sectname,"__la_symbol_ptr")
4777 || !strcmp(sections[i].sectname,"__la_sym_ptr2")
4778 || !strcmp(sections[i].sectname,"__la_sym_ptr3"))
4780 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4783 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr")
4784 || !strcmp(sections[i].sectname,"__pointers"))
4786 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4789 else if(!strcmp(sections[i].sectname,"__jump_table"))
4791 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4797 for(i=0;i<segLC->nsects;i++)
4799 if(!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
4803 #if defined (powerpc_HOST_ARCH)
4804 ocFlushInstructionCache( oc );
4810 #ifdef powerpc_HOST_ARCH
4812 * The Mach-O object format uses leading underscores. But not everywhere.
4813 * There is a small number of runtime support functions defined in
4814 * libcc_dynamic.a whose name does not have a leading underscore.
4815 * As a consequence, we can't get their address from C code.
4816 * We have to use inline assembler just to take the address of a function.
4820 extern void* symbolsWithoutUnderscore[];
4822 static void machoInitSymbolsWithoutUnderscore()
4824 void **p = symbolsWithoutUnderscore;
4825 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
4827 #undef SymI_NeedsProto
4828 #define SymI_NeedsProto(x) \
4829 __asm__ volatile(".long " # x);
4831 RTS_MACHO_NOUNDERLINE_SYMBOLS
4833 __asm__ volatile(".text");
4835 #undef SymI_NeedsProto
4836 #define SymI_NeedsProto(x) \
4837 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
4839 RTS_MACHO_NOUNDERLINE_SYMBOLS
4841 #undef SymI_NeedsProto
4847 * Figure out by how much to shift the entire Mach-O file in memory
4848 * when loading so that its single segment ends up 16-byte-aligned
4850 static int machoGetMisalignment( FILE * f )
4852 struct mach_header header;
4855 fread(&header, sizeof(header), 1, f);
4858 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
4859 if(header.magic != MH_MAGIC_64) {
4860 errorBelch("Bad magic. Expected: %08x, got: %08x.\n",
4861 MH_MAGIC_64, header->magic);
4865 if(header.magic != MH_MAGIC) {
4866 errorBelch("Bad magic. Expected: %08x, got: %08x.\n",
4867 MH_MAGIC, header->magic);
4872 misalignment = (header.sizeofcmds + sizeof(header))
4875 return misalignment ? (16 - misalignment) : 0;