1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team, 2000-2004
7 * ---------------------------------------------------------------------------*/
10 #include "PosixSource.h"
13 /* Linux needs _GNU_SOURCE to get RTLD_DEFAULT from <dlfcn.h> and
14 MREMAP_MAYMOVE from <sys/mman.h>.
25 #include "LinkerInternals.h"
29 #include "RtsTypeable.h"
33 #ifdef HAVE_SYS_TYPES_H
34 #include <sys/types.h>
40 #ifdef HAVE_SYS_STAT_H
44 #if defined(HAVE_DLFCN_H)
48 #if defined(cygwin32_HOST_OS)
53 #ifdef HAVE_SYS_TIME_H
57 #include <sys/fcntl.h>
58 #include <sys/termios.h>
59 #include <sys/utime.h>
60 #include <sys/utsname.h>
64 #if defined(ia64_HOST_ARCH) || defined(linux_HOST_OS) || defined(freebsd_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS)
69 #if defined(linux_HOST_OS) || defined(freebsd_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS)
77 #if defined(linux_HOST_OS) || defined(solaris2_HOST_OS) || defined(freebsd_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS)
78 # define OBJFORMAT_ELF
79 #elif defined(cygwin32_HOST_OS) || defined (mingw32_HOST_OS)
80 # define OBJFORMAT_PEi386
83 #elif defined(darwin_HOST_OS)
84 # define OBJFORMAT_MACHO
85 # include <mach-o/loader.h>
86 # include <mach-o/nlist.h>
87 # include <mach-o/reloc.h>
88 #if !defined(HAVE_DLFCN_H)
89 # include <mach-o/dyld.h>
91 #if defined(powerpc_HOST_ARCH)
92 # include <mach-o/ppc/reloc.h>
94 #if defined(x86_64_HOST_ARCH)
95 # include <mach-o/x86_64/reloc.h>
99 /* Hash table mapping symbol names to Symbol */
100 static /*Str*/HashTable *symhash;
102 /* Hash table mapping symbol names to StgStablePtr */
103 static /*Str*/HashTable *stablehash;
105 /* List of currently loaded objects */
106 ObjectCode *objects = NULL; /* initially empty */
108 #if defined(OBJFORMAT_ELF)
109 static int ocVerifyImage_ELF ( ObjectCode* oc );
110 static int ocGetNames_ELF ( ObjectCode* oc );
111 static int ocResolve_ELF ( ObjectCode* oc );
112 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
113 static int ocAllocateSymbolExtras_ELF ( ObjectCode* oc );
115 #elif defined(OBJFORMAT_PEi386)
116 static int ocVerifyImage_PEi386 ( ObjectCode* oc );
117 static int ocGetNames_PEi386 ( ObjectCode* oc );
118 static int ocResolve_PEi386 ( ObjectCode* oc );
119 static void *lookupSymbolInDLLs ( unsigned char *lbl );
120 static void zapTrailingAtSign ( unsigned char *sym );
121 #elif defined(OBJFORMAT_MACHO)
122 static int ocVerifyImage_MachO ( ObjectCode* oc );
123 static int ocGetNames_MachO ( ObjectCode* oc );
124 static int ocResolve_MachO ( ObjectCode* oc );
126 static int machoGetMisalignment( FILE * );
127 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
128 static int ocAllocateSymbolExtras_MachO ( ObjectCode* oc );
130 #ifdef powerpc_HOST_ARCH
131 static void machoInitSymbolsWithoutUnderscore( void );
135 /* on x86_64 we have a problem with relocating symbol references in
136 * code that was compiled without -fPIC. By default, the small memory
137 * model is used, which assumes that symbol references can fit in a
138 * 32-bit slot. The system dynamic linker makes this work for
139 * references to shared libraries by either (a) allocating a jump
140 * table slot for code references, or (b) moving the symbol at load
141 * time (and copying its contents, if necessary) for data references.
143 * We unfortunately can't tell whether symbol references are to code
144 * or data. So for now we assume they are code (the vast majority
145 * are), and allocate jump-table slots. Unfortunately this will
146 * SILENTLY generate crashing code for data references. This hack is
147 * enabled by X86_64_ELF_NONPIC_HACK.
149 * One workaround is to use shared Haskell libraries. This is
150 * coming. Another workaround is to keep the static libraries but
151 * compile them with -fPIC, because that will generate PIC references
152 * to data which can be relocated. The PIC code is still too green to
153 * do this systematically, though.
156 * See thread http://www.haskell.org/pipermail/cvs-ghc/2007-September/038458.html
158 * Naming Scheme for Symbol Macros
160 * SymI_*: symbol is internal to the RTS. It resides in an object
161 * file/library that is statically.
162 * SymE_*: symbol is external to the RTS library. It might be linked
165 * Sym*_HasProto : the symbol prototype is imported in an include file
166 * or defined explicitly
167 * Sym*_NeedsProto: the symbol is undefined and we add a dummy
168 * default proto extern void sym(void);
170 #define X86_64_ELF_NONPIC_HACK 1
172 /* Link objects into the lower 2Gb on x86_64. GHC assumes the
173 * small memory model on this architecture (see gcc docs,
176 * MAP_32BIT not available on OpenBSD/amd64
178 #if defined(x86_64_HOST_ARCH) && defined(MAP_32BIT)
179 #define TRY_MAP_32BIT MAP_32BIT
181 #define TRY_MAP_32BIT 0
185 * Due to the small memory model (see above), on x86_64 we have to map
186 * all our non-PIC object files into the low 2Gb of the address space
187 * (why 2Gb and not 4Gb? Because all addresses must be reachable
188 * using a 32-bit signed PC-relative offset). On Linux we can do this
189 * using the MAP_32BIT flag to mmap(), however on other OSs
190 * (e.g. *BSD, see #2063, and also on Linux inside Xen, see #2512), we
191 * can't do this. So on these systems, we have to pick a base address
192 * in the low 2Gb of the address space and try to allocate memory from
195 * We pick a default address based on the OS, but also make this
196 * configurable via an RTS flag (+RTS -xm)
198 #if defined(x86_64_HOST_ARCH)
200 #if defined(MAP_32BIT)
201 // Try to use MAP_32BIT
202 #define MMAP_32BIT_BASE_DEFAULT 0
205 #define MMAP_32BIT_BASE_DEFAULT 0x40000000
208 static void *mmap_32bit_base = MMAP_32BIT_BASE_DEFAULT;
211 /* MAP_ANONYMOUS is MAP_ANON on some systems, e.g. OpenBSD */
212 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
213 #define MAP_ANONYMOUS MAP_ANON
216 /* -----------------------------------------------------------------------------
217 * Built-in symbols from the RTS
220 typedef struct _RtsSymbolVal {
226 #define Maybe_Stable_Names SymI_HasProto(mkWeakzh_fast) \
227 SymI_HasProto(mkWeakForeignEnvzh_fast) \
228 SymI_HasProto(makeStableNamezh_fast) \
229 SymI_HasProto(finalizzeWeakzh_fast)
231 /* These are not available in GUM!!! -- HWL */
232 #define Maybe_Stable_Names
235 #if !defined (mingw32_HOST_OS)
236 #define RTS_POSIX_ONLY_SYMBOLS \
237 SymI_HasProto(shutdownHaskellAndSignal) \
238 SymI_NeedsProto(lockFile) \
239 SymI_NeedsProto(unlockFile) \
240 SymI_HasProto(signal_handlers) \
241 SymI_HasProto(stg_sig_install) \
242 SymI_NeedsProto(nocldstop)
245 #if defined (cygwin32_HOST_OS)
246 #define RTS_MINGW_ONLY_SYMBOLS /**/
247 /* Don't have the ability to read import libs / archives, so
248 * we have to stupidly list a lot of what libcygwin.a
251 #define RTS_CYGWIN_ONLY_SYMBOLS \
252 SymI_HasProto(regfree) \
253 SymI_HasProto(regexec) \
254 SymI_HasProto(regerror) \
255 SymI_HasProto(regcomp) \
256 SymI_HasProto(__errno) \
257 SymI_HasProto(access) \
258 SymI_HasProto(chmod) \
259 SymI_HasProto(chdir) \
260 SymI_HasProto(close) \
261 SymI_HasProto(creat) \
263 SymI_HasProto(dup2) \
264 SymI_HasProto(fstat) \
265 SymI_HasProto(fcntl) \
266 SymI_HasProto(getcwd) \
267 SymI_HasProto(getenv) \
268 SymI_HasProto(lseek) \
269 SymI_HasProto(open) \
270 SymI_HasProto(fpathconf) \
271 SymI_HasProto(pathconf) \
272 SymI_HasProto(stat) \
274 SymI_HasProto(tanh) \
275 SymI_HasProto(cosh) \
276 SymI_HasProto(sinh) \
277 SymI_HasProto(atan) \
278 SymI_HasProto(acos) \
279 SymI_HasProto(asin) \
285 SymI_HasProto(sqrt) \
286 SymI_HasProto(localtime_r) \
287 SymI_HasProto(gmtime_r) \
288 SymI_HasProto(mktime) \
289 SymI_NeedsProto(_imp___tzname) \
290 SymI_HasProto(gettimeofday) \
291 SymI_HasProto(timezone) \
292 SymI_HasProto(tcgetattr) \
293 SymI_HasProto(tcsetattr) \
294 SymI_HasProto(memcpy) \
295 SymI_HasProto(memmove) \
296 SymI_HasProto(realloc) \
297 SymI_HasProto(malloc) \
298 SymI_HasProto(free) \
299 SymI_HasProto(fork) \
300 SymI_HasProto(lstat) \
301 SymI_HasProto(isatty) \
302 SymI_HasProto(mkdir) \
303 SymI_HasProto(opendir) \
304 SymI_HasProto(readdir) \
305 SymI_HasProto(rewinddir) \
306 SymI_HasProto(closedir) \
307 SymI_HasProto(link) \
308 SymI_HasProto(mkfifo) \
309 SymI_HasProto(pipe) \
310 SymI_HasProto(read) \
311 SymI_HasProto(rename) \
312 SymI_HasProto(rmdir) \
313 SymI_HasProto(select) \
314 SymI_HasProto(system) \
315 SymI_HasProto(write) \
316 SymI_HasProto(strcmp) \
317 SymI_HasProto(strcpy) \
318 SymI_HasProto(strncpy) \
319 SymI_HasProto(strerror) \
320 SymI_HasProto(sigaddset) \
321 SymI_HasProto(sigemptyset) \
322 SymI_HasProto(sigprocmask) \
323 SymI_HasProto(umask) \
324 SymI_HasProto(uname) \
325 SymI_HasProto(unlink) \
326 SymI_HasProto(utime) \
327 SymI_HasProto(waitpid)
329 #elif !defined(mingw32_HOST_OS)
330 #define RTS_MINGW_ONLY_SYMBOLS /**/
331 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
332 #else /* defined(mingw32_HOST_OS) */
333 #define RTS_POSIX_ONLY_SYMBOLS /**/
334 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
336 /* Extra syms gen'ed by mingw-2's gcc-3.2: */
338 #define RTS_MINGW_EXTRA_SYMS \
339 SymI_NeedsProto(_imp____mb_cur_max) \
340 SymI_NeedsProto(_imp___pctype)
342 #define RTS_MINGW_EXTRA_SYMS
345 #if HAVE_GETTIMEOFDAY
346 #define RTS_MINGW_GETTIMEOFDAY_SYM SymI_NeedsProto(gettimeofday)
348 #define RTS_MINGW_GETTIMEOFDAY_SYM /**/
351 /* These are statically linked from the mingw libraries into the ghc
352 executable, so we have to employ this hack. */
353 #define RTS_MINGW_ONLY_SYMBOLS \
354 SymI_HasProto(asyncReadzh_fast) \
355 SymI_HasProto(asyncWritezh_fast) \
356 SymI_HasProto(asyncDoProczh_fast) \
357 SymI_HasProto(memset) \
358 SymI_HasProto(inet_ntoa) \
359 SymI_HasProto(inet_addr) \
360 SymI_HasProto(htonl) \
361 SymI_HasProto(recvfrom) \
362 SymI_HasProto(listen) \
363 SymI_HasProto(bind) \
364 SymI_HasProto(shutdown) \
365 SymI_HasProto(connect) \
366 SymI_HasProto(htons) \
367 SymI_HasProto(ntohs) \
368 SymI_HasProto(getservbyname) \
369 SymI_HasProto(getservbyport) \
370 SymI_HasProto(getprotobynumber) \
371 SymI_HasProto(getprotobyname) \
372 SymI_HasProto(gethostbyname) \
373 SymI_HasProto(gethostbyaddr) \
374 SymI_HasProto(gethostname) \
375 SymI_HasProto(strcpy) \
376 SymI_HasProto(strncpy) \
377 SymI_HasProto(abort) \
378 SymI_NeedsProto(_alloca) \
379 SymI_NeedsProto(isxdigit) \
380 SymI_NeedsProto(isupper) \
381 SymI_NeedsProto(ispunct) \
382 SymI_NeedsProto(islower) \
383 SymI_NeedsProto(isspace) \
384 SymI_NeedsProto(isprint) \
385 SymI_NeedsProto(isdigit) \
386 SymI_NeedsProto(iscntrl) \
387 SymI_NeedsProto(isalpha) \
388 SymI_NeedsProto(isalnum) \
389 SymI_HasProto(strcmp) \
390 SymI_HasProto(memmove) \
391 SymI_HasProto(realloc) \
392 SymI_HasProto(malloc) \
394 SymI_HasProto(tanh) \
395 SymI_HasProto(cosh) \
396 SymI_HasProto(sinh) \
397 SymI_HasProto(atan) \
398 SymI_HasProto(acos) \
399 SymI_HasProto(asin) \
405 SymI_HasProto(sqrt) \
406 SymI_HasProto(powf) \
407 SymI_HasProto(tanhf) \
408 SymI_HasProto(coshf) \
409 SymI_HasProto(sinhf) \
410 SymI_HasProto(atanf) \
411 SymI_HasProto(acosf) \
412 SymI_HasProto(asinf) \
413 SymI_HasProto(tanf) \
414 SymI_HasProto(cosf) \
415 SymI_HasProto(sinf) \
416 SymI_HasProto(expf) \
417 SymI_HasProto(logf) \
418 SymI_HasProto(sqrtf) \
419 SymI_HasProto(memcpy) \
420 SymI_HasProto(rts_InstallConsoleEvent) \
421 SymI_HasProto(rts_ConsoleHandlerDone) \
422 SymI_NeedsProto(mktime) \
423 SymI_NeedsProto(_imp___timezone) \
424 SymI_NeedsProto(_imp___tzname) \
425 SymI_NeedsProto(_imp__tzname) \
426 SymI_NeedsProto(_imp___iob) \
427 SymI_NeedsProto(_imp___osver) \
428 SymI_NeedsProto(localtime) \
429 SymI_NeedsProto(gmtime) \
430 SymI_NeedsProto(opendir) \
431 SymI_NeedsProto(readdir) \
432 SymI_NeedsProto(rewinddir) \
433 RTS_MINGW_EXTRA_SYMS \
434 RTS_MINGW_GETTIMEOFDAY_SYM \
435 SymI_NeedsProto(closedir)
438 #if defined(darwin_TARGET_OS) && HAVE_PRINTF_LDBLSTUB
439 #define RTS_DARWIN_ONLY_SYMBOLS \
440 SymI_NeedsProto(asprintf$LDBLStub) \
441 SymI_NeedsProto(err$LDBLStub) \
442 SymI_NeedsProto(errc$LDBLStub) \
443 SymI_NeedsProto(errx$LDBLStub) \
444 SymI_NeedsProto(fprintf$LDBLStub) \
445 SymI_NeedsProto(fscanf$LDBLStub) \
446 SymI_NeedsProto(fwprintf$LDBLStub) \
447 SymI_NeedsProto(fwscanf$LDBLStub) \
448 SymI_NeedsProto(printf$LDBLStub) \
449 SymI_NeedsProto(scanf$LDBLStub) \
450 SymI_NeedsProto(snprintf$LDBLStub) \
451 SymI_NeedsProto(sprintf$LDBLStub) \
452 SymI_NeedsProto(sscanf$LDBLStub) \
453 SymI_NeedsProto(strtold$LDBLStub) \
454 SymI_NeedsProto(swprintf$LDBLStub) \
455 SymI_NeedsProto(swscanf$LDBLStub) \
456 SymI_NeedsProto(syslog$LDBLStub) \
457 SymI_NeedsProto(vasprintf$LDBLStub) \
458 SymI_NeedsProto(verr$LDBLStub) \
459 SymI_NeedsProto(verrc$LDBLStub) \
460 SymI_NeedsProto(verrx$LDBLStub) \
461 SymI_NeedsProto(vfprintf$LDBLStub) \
462 SymI_NeedsProto(vfscanf$LDBLStub) \
463 SymI_NeedsProto(vfwprintf$LDBLStub) \
464 SymI_NeedsProto(vfwscanf$LDBLStub) \
465 SymI_NeedsProto(vprintf$LDBLStub) \
466 SymI_NeedsProto(vscanf$LDBLStub) \
467 SymI_NeedsProto(vsnprintf$LDBLStub) \
468 SymI_NeedsProto(vsprintf$LDBLStub) \
469 SymI_NeedsProto(vsscanf$LDBLStub) \
470 SymI_NeedsProto(vswprintf$LDBLStub) \
471 SymI_NeedsProto(vswscanf$LDBLStub) \
472 SymI_NeedsProto(vsyslog$LDBLStub) \
473 SymI_NeedsProto(vwarn$LDBLStub) \
474 SymI_NeedsProto(vwarnc$LDBLStub) \
475 SymI_NeedsProto(vwarnx$LDBLStub) \
476 SymI_NeedsProto(vwprintf$LDBLStub) \
477 SymI_NeedsProto(vwscanf$LDBLStub) \
478 SymI_NeedsProto(warn$LDBLStub) \
479 SymI_NeedsProto(warnc$LDBLStub) \
480 SymI_NeedsProto(warnx$LDBLStub) \
481 SymI_NeedsProto(wcstold$LDBLStub) \
482 SymI_NeedsProto(wprintf$LDBLStub) \
483 SymI_NeedsProto(wscanf$LDBLStub)
485 #define RTS_DARWIN_ONLY_SYMBOLS
489 # define MAIN_CAP_SYM SymI_HasProto(MainCapability)
491 # define MAIN_CAP_SYM
494 #if !defined(mingw32_HOST_OS)
495 #define RTS_USER_SIGNALS_SYMBOLS \
496 SymI_HasProto(setIOManagerPipe)
498 #define RTS_USER_SIGNALS_SYMBOLS \
499 SymI_HasProto(sendIOManagerEvent) \
500 SymI_HasProto(readIOManagerEvent) \
501 SymI_HasProto(getIOManagerEvent) \
502 SymI_HasProto(console_handler)
505 #define RTS_LIBFFI_SYMBOLS \
506 SymE_NeedsProto(ffi_prep_cif) \
507 SymE_NeedsProto(ffi_call) \
508 SymE_NeedsProto(ffi_type_void) \
509 SymE_NeedsProto(ffi_type_float) \
510 SymE_NeedsProto(ffi_type_double) \
511 SymE_NeedsProto(ffi_type_sint64) \
512 SymE_NeedsProto(ffi_type_uint64) \
513 SymE_NeedsProto(ffi_type_sint32) \
514 SymE_NeedsProto(ffi_type_uint32) \
515 SymE_NeedsProto(ffi_type_sint16) \
516 SymE_NeedsProto(ffi_type_uint16) \
517 SymE_NeedsProto(ffi_type_sint8) \
518 SymE_NeedsProto(ffi_type_uint8) \
519 SymE_NeedsProto(ffi_type_pointer)
521 #ifdef TABLES_NEXT_TO_CODE
522 #define RTS_RET_SYMBOLS /* nothing */
524 #define RTS_RET_SYMBOLS \
525 SymI_HasProto(stg_enter_ret) \
526 SymI_HasProto(stg_gc_fun_ret) \
527 SymI_HasProto(stg_ap_v_ret) \
528 SymI_HasProto(stg_ap_f_ret) \
529 SymI_HasProto(stg_ap_d_ret) \
530 SymI_HasProto(stg_ap_l_ret) \
531 SymI_HasProto(stg_ap_n_ret) \
532 SymI_HasProto(stg_ap_p_ret) \
533 SymI_HasProto(stg_ap_pv_ret) \
534 SymI_HasProto(stg_ap_pp_ret) \
535 SymI_HasProto(stg_ap_ppv_ret) \
536 SymI_HasProto(stg_ap_ppp_ret) \
537 SymI_HasProto(stg_ap_pppv_ret) \
538 SymI_HasProto(stg_ap_pppp_ret) \
539 SymI_HasProto(stg_ap_ppppp_ret) \
540 SymI_HasProto(stg_ap_pppppp_ret)
543 /* On Windows, we link libgmp.a statically into libHSrts.dll */
544 #ifdef mingw32_HOST_OS
546 SymI_HasProto(__gmpz_cmp) \
547 SymI_HasProto(__gmpz_cmp_si) \
548 SymI_HasProto(__gmpz_cmp_ui) \
549 SymI_HasProto(__gmpz_get_si) \
550 SymI_HasProto(__gmpz_get_ui)
553 SymE_HasProto(__gmpz_cmp) \
554 SymE_HasProto(__gmpz_cmp_si) \
555 SymE_HasProto(__gmpz_cmp_ui) \
556 SymE_HasProto(__gmpz_get_si) \
557 SymE_HasProto(__gmpz_get_ui)
560 #define RTS_SYMBOLS \
562 SymI_HasProto(StgReturn) \
563 SymI_HasProto(stg_enter_info) \
564 SymI_HasProto(stg_gc_void_info) \
565 SymI_HasProto(__stg_gc_enter_1) \
566 SymI_HasProto(stg_gc_noregs) \
567 SymI_HasProto(stg_gc_unpt_r1_info) \
568 SymI_HasProto(stg_gc_unpt_r1) \
569 SymI_HasProto(stg_gc_unbx_r1_info) \
570 SymI_HasProto(stg_gc_unbx_r1) \
571 SymI_HasProto(stg_gc_f1_info) \
572 SymI_HasProto(stg_gc_f1) \
573 SymI_HasProto(stg_gc_d1_info) \
574 SymI_HasProto(stg_gc_d1) \
575 SymI_HasProto(stg_gc_l1_info) \
576 SymI_HasProto(stg_gc_l1) \
577 SymI_HasProto(__stg_gc_fun) \
578 SymI_HasProto(stg_gc_fun_info) \
579 SymI_HasProto(stg_gc_gen) \
580 SymI_HasProto(stg_gc_gen_info) \
581 SymI_HasProto(stg_gc_gen_hp) \
582 SymI_HasProto(stg_gc_ut) \
583 SymI_HasProto(stg_gen_yield) \
584 SymI_HasProto(stg_yield_noregs) \
585 SymI_HasProto(stg_yield_to_interpreter) \
586 SymI_HasProto(stg_gen_block) \
587 SymI_HasProto(stg_block_noregs) \
588 SymI_HasProto(stg_block_1) \
589 SymI_HasProto(stg_block_takemvar) \
590 SymI_HasProto(stg_block_putmvar) \
592 SymI_HasProto(MallocFailHook) \
593 SymI_HasProto(OnExitHook) \
594 SymI_HasProto(OutOfHeapHook) \
595 SymI_HasProto(StackOverflowHook) \
596 SymI_HasProto(__encodeDouble) \
597 SymI_HasProto(__encodeFloat) \
598 SymI_HasProto(addDLL) \
600 SymI_HasProto(__int_encodeDouble) \
601 SymI_HasProto(__word_encodeDouble) \
602 SymI_HasProto(__2Int_encodeDouble) \
603 SymI_HasProto(__int_encodeFloat) \
604 SymI_HasProto(__word_encodeFloat) \
605 SymI_HasProto(andIntegerzh_fast) \
606 SymI_HasProto(atomicallyzh_fast) \
607 SymI_HasProto(barf) \
608 SymI_HasProto(debugBelch) \
609 SymI_HasProto(errorBelch) \
610 SymI_HasProto(asyncExceptionsBlockedzh_fast) \
611 SymI_HasProto(blockAsyncExceptionszh_fast) \
612 SymI_HasProto(catchzh_fast) \
613 SymI_HasProto(catchRetryzh_fast) \
614 SymI_HasProto(catchSTMzh_fast) \
615 SymI_HasProto(checkzh_fast) \
616 SymI_HasProto(closure_flags) \
617 SymI_HasProto(cmp_thread) \
618 SymI_HasProto(cmpIntegerzh_fast) \
619 SymI_HasProto(cmpIntegerIntzh_fast) \
620 SymI_HasProto(complementIntegerzh_fast) \
621 SymI_HasProto(createAdjustor) \
622 SymI_HasProto(decodeDoublezh_fast) \
623 SymI_HasProto(decodeFloatzh_fast) \
624 SymI_HasProto(decodeDoublezu2Intzh_fast) \
625 SymI_HasProto(decodeFloatzuIntzh_fast) \
626 SymI_HasProto(defaultsHook) \
627 SymI_HasProto(delayzh_fast) \
628 SymI_HasProto(deRefWeakzh_fast) \
629 SymI_HasProto(deRefStablePtrzh_fast) \
630 SymI_HasProto(dirty_MUT_VAR) \
631 SymI_HasProto(divExactIntegerzh_fast) \
632 SymI_HasProto(divModIntegerzh_fast) \
633 SymI_HasProto(forkzh_fast) \
634 SymI_HasProto(forkOnzh_fast) \
635 SymI_HasProto(forkProcess) \
636 SymI_HasProto(forkOS_createThread) \
637 SymI_HasProto(freeHaskellFunctionPtr) \
638 SymI_HasProto(freeStablePtr) \
639 SymI_HasProto(getOrSetTypeableStore) \
640 SymI_HasProto(gcdIntegerzh_fast) \
641 SymI_HasProto(gcdIntegerIntzh_fast) \
642 SymI_HasProto(gcdIntzh_fast) \
643 SymI_HasProto(genSymZh) \
644 SymI_HasProto(genericRaise) \
645 SymI_HasProto(getProgArgv) \
646 SymI_HasProto(getFullProgArgv) \
647 SymI_HasProto(getStablePtr) \
648 SymI_HasProto(hs_init) \
649 SymI_HasProto(hs_exit) \
650 SymI_HasProto(hs_set_argv) \
651 SymI_HasProto(hs_add_root) \
652 SymI_HasProto(hs_perform_gc) \
653 SymI_HasProto(hs_free_stable_ptr) \
654 SymI_HasProto(hs_free_fun_ptr) \
655 SymI_HasProto(hs_hpc_rootModule) \
656 SymI_HasProto(initLinker) \
657 SymI_HasProto(unpackClosurezh_fast) \
658 SymI_HasProto(getApStackValzh_fast) \
659 SymI_HasProto(getSparkzh_fast) \
660 SymI_HasProto(int2Integerzh_fast) \
661 SymI_HasProto(integer2Intzh_fast) \
662 SymI_HasProto(integer2Wordzh_fast) \
663 SymI_HasProto(isCurrentThreadBoundzh_fast) \
664 SymI_HasProto(isDoubleDenormalized) \
665 SymI_HasProto(isDoubleInfinite) \
666 SymI_HasProto(isDoubleNaN) \
667 SymI_HasProto(isDoubleNegativeZero) \
668 SymI_HasProto(isEmptyMVarzh_fast) \
669 SymI_HasProto(isFloatDenormalized) \
670 SymI_HasProto(isFloatInfinite) \
671 SymI_HasProto(isFloatNaN) \
672 SymI_HasProto(isFloatNegativeZero) \
673 SymI_HasProto(killThreadzh_fast) \
674 SymI_HasProto(loadObj) \
675 SymI_HasProto(insertStableSymbol) \
676 SymI_HasProto(insertSymbol) \
677 SymI_HasProto(lookupSymbol) \
678 SymI_HasProto(makeStablePtrzh_fast) \
679 SymI_HasProto(minusIntegerzh_fast) \
680 SymI_HasProto(mkApUpd0zh_fast) \
681 SymI_HasProto(myThreadIdzh_fast) \
682 SymI_HasProto(labelThreadzh_fast) \
683 SymI_HasProto(newArrayzh_fast) \
684 SymI_HasProto(newBCOzh_fast) \
685 SymI_HasProto(newByteArrayzh_fast) \
686 SymI_HasProto_redirect(newCAF, newDynCAF) \
687 SymI_HasProto(newMVarzh_fast) \
688 SymI_HasProto(newMutVarzh_fast) \
689 SymI_HasProto(newTVarzh_fast) \
690 SymI_HasProto(noDuplicatezh_fast) \
691 SymI_HasProto(atomicModifyMutVarzh_fast) \
692 SymI_HasProto(newPinnedByteArrayzh_fast) \
693 SymI_HasProto(newSpark) \
694 SymI_HasProto(orIntegerzh_fast) \
695 SymI_HasProto(performGC) \
696 SymI_HasProto(performMajorGC) \
697 SymI_HasProto(plusIntegerzh_fast) \
698 SymI_HasProto(prog_argc) \
699 SymI_HasProto(prog_argv) \
700 SymI_HasProto(putMVarzh_fast) \
701 SymI_HasProto(quotIntegerzh_fast) \
702 SymI_HasProto(quotRemIntegerzh_fast) \
703 SymI_HasProto(raisezh_fast) \
704 SymI_HasProto(raiseIOzh_fast) \
705 SymI_HasProto(readTVarzh_fast) \
706 SymI_HasProto(readTVarIOzh_fast) \
707 SymI_HasProto(remIntegerzh_fast) \
708 SymI_HasProto(resetNonBlockingFd) \
709 SymI_HasProto(resumeThread) \
710 SymI_HasProto(resolveObjs) \
711 SymI_HasProto(retryzh_fast) \
712 SymI_HasProto(rts_apply) \
713 SymI_HasProto(rts_checkSchedStatus) \
714 SymI_HasProto(rts_eval) \
715 SymI_HasProto(rts_evalIO) \
716 SymI_HasProto(rts_evalLazyIO) \
717 SymI_HasProto(rts_evalStableIO) \
718 SymI_HasProto(rts_eval_) \
719 SymI_HasProto(rts_getBool) \
720 SymI_HasProto(rts_getChar) \
721 SymI_HasProto(rts_getDouble) \
722 SymI_HasProto(rts_getFloat) \
723 SymI_HasProto(rts_getInt) \
724 SymI_HasProto(rts_getInt8) \
725 SymI_HasProto(rts_getInt16) \
726 SymI_HasProto(rts_getInt32) \
727 SymI_HasProto(rts_getInt64) \
728 SymI_HasProto(rts_getPtr) \
729 SymI_HasProto(rts_getFunPtr) \
730 SymI_HasProto(rts_getStablePtr) \
731 SymI_HasProto(rts_getThreadId) \
732 SymI_HasProto(rts_getWord) \
733 SymI_HasProto(rts_getWord8) \
734 SymI_HasProto(rts_getWord16) \
735 SymI_HasProto(rts_getWord32) \
736 SymI_HasProto(rts_getWord64) \
737 SymI_HasProto(rts_lock) \
738 SymI_HasProto(rts_mkBool) \
739 SymI_HasProto(rts_mkChar) \
740 SymI_HasProto(rts_mkDouble) \
741 SymI_HasProto(rts_mkFloat) \
742 SymI_HasProto(rts_mkInt) \
743 SymI_HasProto(rts_mkInt8) \
744 SymI_HasProto(rts_mkInt16) \
745 SymI_HasProto(rts_mkInt32) \
746 SymI_HasProto(rts_mkInt64) \
747 SymI_HasProto(rts_mkPtr) \
748 SymI_HasProto(rts_mkFunPtr) \
749 SymI_HasProto(rts_mkStablePtr) \
750 SymI_HasProto(rts_mkString) \
751 SymI_HasProto(rts_mkWord) \
752 SymI_HasProto(rts_mkWord8) \
753 SymI_HasProto(rts_mkWord16) \
754 SymI_HasProto(rts_mkWord32) \
755 SymI_HasProto(rts_mkWord64) \
756 SymI_HasProto(rts_unlock) \
757 SymI_HasProto(rtsSupportsBoundThreads) \
758 SymI_HasProto(__hscore_get_saved_termios) \
759 SymI_HasProto(__hscore_set_saved_termios) \
760 SymI_HasProto(setProgArgv) \
761 SymI_HasProto(startupHaskell) \
762 SymI_HasProto(shutdownHaskell) \
763 SymI_HasProto(shutdownHaskellAndExit) \
764 SymI_HasProto(stable_ptr_table) \
765 SymI_HasProto(stackOverflow) \
766 SymI_HasProto(stg_CAF_BLACKHOLE_info) \
767 SymI_HasProto(__stg_EAGER_BLACKHOLE_info) \
768 SymI_HasProto(awakenBlockedQueue) \
769 SymI_HasProto(startTimer) \
770 SymI_HasProto(stg_CHARLIKE_closure) \
771 SymI_HasProto(stg_MVAR_CLEAN_info) \
772 SymI_HasProto(stg_MVAR_DIRTY_info) \
773 SymI_HasProto(stg_IND_STATIC_info) \
774 SymI_HasProto(stg_INTLIKE_closure) \
775 SymI_HasProto(stg_MUT_ARR_PTRS_DIRTY_info) \
776 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN_info) \
777 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN0_info) \
778 SymI_HasProto(stg_WEAK_info) \
779 SymI_HasProto(stg_ap_v_info) \
780 SymI_HasProto(stg_ap_f_info) \
781 SymI_HasProto(stg_ap_d_info) \
782 SymI_HasProto(stg_ap_l_info) \
783 SymI_HasProto(stg_ap_n_info) \
784 SymI_HasProto(stg_ap_p_info) \
785 SymI_HasProto(stg_ap_pv_info) \
786 SymI_HasProto(stg_ap_pp_info) \
787 SymI_HasProto(stg_ap_ppv_info) \
788 SymI_HasProto(stg_ap_ppp_info) \
789 SymI_HasProto(stg_ap_pppv_info) \
790 SymI_HasProto(stg_ap_pppp_info) \
791 SymI_HasProto(stg_ap_ppppp_info) \
792 SymI_HasProto(stg_ap_pppppp_info) \
793 SymI_HasProto(stg_ap_0_fast) \
794 SymI_HasProto(stg_ap_v_fast) \
795 SymI_HasProto(stg_ap_f_fast) \
796 SymI_HasProto(stg_ap_d_fast) \
797 SymI_HasProto(stg_ap_l_fast) \
798 SymI_HasProto(stg_ap_n_fast) \
799 SymI_HasProto(stg_ap_p_fast) \
800 SymI_HasProto(stg_ap_pv_fast) \
801 SymI_HasProto(stg_ap_pp_fast) \
802 SymI_HasProto(stg_ap_ppv_fast) \
803 SymI_HasProto(stg_ap_ppp_fast) \
804 SymI_HasProto(stg_ap_pppv_fast) \
805 SymI_HasProto(stg_ap_pppp_fast) \
806 SymI_HasProto(stg_ap_ppppp_fast) \
807 SymI_HasProto(stg_ap_pppppp_fast) \
808 SymI_HasProto(stg_ap_1_upd_info) \
809 SymI_HasProto(stg_ap_2_upd_info) \
810 SymI_HasProto(stg_ap_3_upd_info) \
811 SymI_HasProto(stg_ap_4_upd_info) \
812 SymI_HasProto(stg_ap_5_upd_info) \
813 SymI_HasProto(stg_ap_6_upd_info) \
814 SymI_HasProto(stg_ap_7_upd_info) \
815 SymI_HasProto(stg_exit) \
816 SymI_HasProto(stg_sel_0_upd_info) \
817 SymI_HasProto(stg_sel_10_upd_info) \
818 SymI_HasProto(stg_sel_11_upd_info) \
819 SymI_HasProto(stg_sel_12_upd_info) \
820 SymI_HasProto(stg_sel_13_upd_info) \
821 SymI_HasProto(stg_sel_14_upd_info) \
822 SymI_HasProto(stg_sel_15_upd_info) \
823 SymI_HasProto(stg_sel_1_upd_info) \
824 SymI_HasProto(stg_sel_2_upd_info) \
825 SymI_HasProto(stg_sel_3_upd_info) \
826 SymI_HasProto(stg_sel_4_upd_info) \
827 SymI_HasProto(stg_sel_5_upd_info) \
828 SymI_HasProto(stg_sel_6_upd_info) \
829 SymI_HasProto(stg_sel_7_upd_info) \
830 SymI_HasProto(stg_sel_8_upd_info) \
831 SymI_HasProto(stg_sel_9_upd_info) \
832 SymI_HasProto(stg_upd_frame_info) \
833 SymI_HasProto(suspendThread) \
834 SymI_HasProto(takeMVarzh_fast) \
835 SymI_HasProto(threadStatuszh_fast) \
836 SymI_HasProto(timesIntegerzh_fast) \
837 SymI_HasProto(tryPutMVarzh_fast) \
838 SymI_HasProto(tryTakeMVarzh_fast) \
839 SymI_HasProto(unblockAsyncExceptionszh_fast) \
840 SymI_HasProto(unloadObj) \
841 SymI_HasProto(unsafeThawArrayzh_fast) \
842 SymI_HasProto(waitReadzh_fast) \
843 SymI_HasProto(waitWritezh_fast) \
844 SymI_HasProto(word2Integerzh_fast) \
845 SymI_HasProto(writeTVarzh_fast) \
846 SymI_HasProto(xorIntegerzh_fast) \
847 SymI_HasProto(yieldzh_fast) \
848 SymI_NeedsProto(stg_interp_constr_entry) \
849 SymI_HasProto(allocateExec) \
850 SymI_HasProto(freeExec) \
851 SymI_HasProto(getAllocations) \
852 SymI_HasProto(revertCAFs) \
853 SymI_HasProto(RtsFlags) \
854 SymI_NeedsProto(rts_breakpoint_io_action) \
855 SymI_NeedsProto(rts_stop_next_breakpoint) \
856 SymI_NeedsProto(rts_stop_on_exception) \
857 SymI_HasProto(stopTimer) \
858 SymI_HasProto(n_capabilities) \
859 RTS_USER_SIGNALS_SYMBOLS
861 #ifdef SUPPORT_LONG_LONGS
862 #define RTS_LONG_LONG_SYMS \
863 SymI_HasProto(int64ToIntegerzh_fast) \
864 SymI_HasProto(word64ToIntegerzh_fast)
866 #define RTS_LONG_LONG_SYMS /* nothing */
869 // 64-bit support functions in libgcc.a
870 #if defined(__GNUC__) && SIZEOF_VOID_P <= 4
871 #define RTS_LIBGCC_SYMBOLS \
872 SymI_NeedsProto(__divdi3) \
873 SymI_NeedsProto(__udivdi3) \
874 SymI_NeedsProto(__moddi3) \
875 SymI_NeedsProto(__umoddi3) \
876 SymI_NeedsProto(__muldi3) \
877 SymI_NeedsProto(__ashldi3) \
878 SymI_NeedsProto(__ashrdi3) \
879 SymI_NeedsProto(__lshrdi3) \
880 SymI_NeedsProto(__eprintf)
881 #elif defined(ia64_HOST_ARCH)
882 #define RTS_LIBGCC_SYMBOLS \
883 SymI_NeedsProto(__divdi3) \
884 SymI_NeedsProto(__udivdi3) \
885 SymI_NeedsProto(__moddi3) \
886 SymI_NeedsProto(__umoddi3) \
887 SymI_NeedsProto(__divsf3) \
888 SymI_NeedsProto(__divdf3)
890 #define RTS_LIBGCC_SYMBOLS
893 #if defined(darwin_HOST_OS) && defined(powerpc_HOST_ARCH)
894 // Symbols that don't have a leading underscore
895 // on Mac OS X. They have to receive special treatment,
896 // see machoInitSymbolsWithoutUnderscore()
897 #define RTS_MACHO_NOUNDERLINE_SYMBOLS \
898 SymI_NeedsProto(saveFP) \
899 SymI_NeedsProto(restFP)
902 /* entirely bogus claims about types of these symbols */
903 #define SymI_NeedsProto(vvv) extern void vvv(void);
904 #if defined(__PIC__) && defined(mingw32_TARGET_OS)
905 #define SymE_HasProto(vvv) SymE_HasProto(vvv);
906 #define SymE_NeedsProto(vvv) extern void _imp__ ## vvv (void);
908 #define SymE_NeedsProto(vvv) SymI_NeedsProto(vvv);
909 #define SymE_HasProto(vvv) SymI_HasProto(vvv)
911 #define SymI_HasProto(vvv) /**/
912 #define SymI_HasProto_redirect(vvv,xxx) /**/
916 RTS_POSIX_ONLY_SYMBOLS
917 RTS_MINGW_ONLY_SYMBOLS
918 RTS_CYGWIN_ONLY_SYMBOLS
919 RTS_DARWIN_ONLY_SYMBOLS
922 #undef SymI_NeedsProto
924 #undef SymI_HasProto_redirect
926 #undef SymE_NeedsProto
928 #ifdef LEADING_UNDERSCORE
929 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
931 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
934 #define SymI_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
936 #define SymE_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
937 (void*)DLL_IMPORT_DATA_REF(vvv) },
939 #define SymI_NeedsProto(vvv) SymI_HasProto(vvv)
940 #define SymE_NeedsProto(vvv) SymE_HasProto(vvv)
942 // SymI_HasProto_redirect allows us to redirect references to one symbol to
943 // another symbol. See newCAF/newDynCAF for an example.
944 #define SymI_HasProto_redirect(vvv,xxx) \
945 { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
948 static RtsSymbolVal rtsSyms[] = {
952 RTS_POSIX_ONLY_SYMBOLS
953 RTS_MINGW_ONLY_SYMBOLS
954 RTS_CYGWIN_ONLY_SYMBOLS
955 RTS_DARWIN_ONLY_SYMBOLS
958 #if defined(darwin_HOST_OS) && defined(i386_HOST_ARCH)
959 // dyld stub code contains references to this,
960 // but it should never be called because we treat
961 // lazy pointers as nonlazy.
962 { "dyld_stub_binding_helper", (void*)0xDEADBEEF },
964 { 0, 0 } /* sentinel */
969 /* -----------------------------------------------------------------------------
970 * Insert symbols into hash tables, checking for duplicates.
973 static void ghciInsertStrHashTable ( char* obj_name,
979 if (lookupHashTable(table, (StgWord)key) == NULL)
981 insertStrHashTable(table, (StgWord)key, data);
986 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
988 "whilst processing object file\n"
990 "This could be caused by:\n"
991 " * Loading two different object files which export the same symbol\n"
992 " * Specifying the same object file twice on the GHCi command line\n"
993 " * An incorrect `package.conf' entry, causing some object to be\n"
995 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
1002 /* -----------------------------------------------------------------------------
1003 * initialize the object linker
1007 static int linker_init_done = 0 ;
1009 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1010 static void *dl_prog_handle;
1018 /* Make initLinker idempotent, so we can call it
1019 before evey relevant operation; that means we
1020 don't need to initialise the linker separately */
1021 if (linker_init_done == 1) { return; } else {
1022 linker_init_done = 1;
1025 stablehash = allocStrHashTable();
1026 symhash = allocStrHashTable();
1028 /* populate the symbol table with stuff from the RTS */
1029 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
1030 ghciInsertStrHashTable("(GHCi built-in symbols)",
1031 symhash, sym->lbl, sym->addr);
1033 # if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH)
1034 machoInitSymbolsWithoutUnderscore();
1037 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1038 # if defined(RTLD_DEFAULT)
1039 dl_prog_handle = RTLD_DEFAULT;
1041 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
1042 # endif /* RTLD_DEFAULT */
1045 #if defined(x86_64_HOST_ARCH)
1046 if (RtsFlags.MiscFlags.linkerMemBase != 0) {
1047 // User-override for mmap_32bit_base
1048 mmap_32bit_base = (void*)RtsFlags.MiscFlags.linkerMemBase;
1053 /* -----------------------------------------------------------------------------
1054 * Loading DLL or .so dynamic libraries
1055 * -----------------------------------------------------------------------------
1057 * Add a DLL from which symbols may be found. In the ELF case, just
1058 * do RTLD_GLOBAL-style add, so no further messing around needs to
1059 * happen in order that symbols in the loaded .so are findable --
1060 * lookupSymbol() will subsequently see them by dlsym on the program's
1061 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
1063 * In the PEi386 case, open the DLLs and put handles to them in a
1064 * linked list. When looking for a symbol, try all handles in the
1065 * list. This means that we need to load even DLLs that are guaranteed
1066 * to be in the ghc.exe image already, just so we can get a handle
1067 * to give to loadSymbol, so that we can find the symbols. For such
1068 * libraries, the LoadLibrary call should be a no-op except for returning
1073 #if defined(OBJFORMAT_PEi386)
1074 /* A record for storing handles into DLLs. */
1079 struct _OpenedDLL* next;
1084 /* A list thereof. */
1085 static OpenedDLL* opened_dlls = NULL;
1089 addDLL( char *dll_name )
1091 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1092 /* ------------------- ELF DLL loader ------------------- */
1098 // omitted: RTLD_NOW
1099 // see http://www.haskell.org/pipermail/cvs-ghc/2007-September/038570.html
1100 hdl= dlopen(dll_name, RTLD_LAZY | RTLD_GLOBAL);
1103 /* dlopen failed; return a ptr to the error msg. */
1105 if (errmsg == NULL) errmsg = "addDLL: unknown error";
1112 # elif defined(OBJFORMAT_PEi386)
1113 /* ------------------- Win32 DLL loader ------------------- */
1121 /* debugBelch("\naddDLL; dll_name = `%s'\n", dll_name); */
1123 /* See if we've already got it, and ignore if so. */
1124 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
1125 if (0 == strcmp(o_dll->name, dll_name))
1129 /* The file name has no suffix (yet) so that we can try
1130 both foo.dll and foo.drv
1132 The documentation for LoadLibrary says:
1133 If no file name extension is specified in the lpFileName
1134 parameter, the default library extension .dll is
1135 appended. However, the file name string can include a trailing
1136 point character (.) to indicate that the module name has no
1139 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
1140 sprintf(buf, "%s.DLL", dll_name);
1141 instance = LoadLibrary(buf);
1142 if (instance == NULL) {
1143 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1144 // KAA: allow loading of drivers (like winspool.drv)
1145 sprintf(buf, "%s.DRV", dll_name);
1146 instance = LoadLibrary(buf);
1147 if (instance == NULL) {
1148 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1149 // #1883: allow loading of unix-style libfoo.dll DLLs
1150 sprintf(buf, "lib%s.DLL", dll_name);
1151 instance = LoadLibrary(buf);
1152 if (instance == NULL) {
1159 /* Add this DLL to the list of DLLs in which to search for symbols. */
1160 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
1161 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
1162 strcpy(o_dll->name, dll_name);
1163 o_dll->instance = instance;
1164 o_dll->next = opened_dlls;
1165 opened_dlls = o_dll;
1171 sysErrorBelch(dll_name);
1173 /* LoadLibrary failed; return a ptr to the error msg. */
1174 return "addDLL: could not load DLL";
1177 barf("addDLL: not implemented on this platform");
1181 /* -----------------------------------------------------------------------------
1182 * insert a stable symbol in the hash table
1186 insertStableSymbol(char* obj_name, char* key, StgPtr p)
1188 ghciInsertStrHashTable(obj_name, stablehash, key, getStablePtr(p));
1192 /* -----------------------------------------------------------------------------
1193 * insert a symbol in the hash table
1196 insertSymbol(char* obj_name, char* key, void* data)
1198 ghciInsertStrHashTable(obj_name, symhash, key, data);
1201 /* -----------------------------------------------------------------------------
1202 * lookup a symbol in the hash table
1205 lookupSymbol( char *lbl )
1209 ASSERT(symhash != NULL);
1210 val = lookupStrHashTable(symhash, lbl);
1213 # if defined(OBJFORMAT_ELF)
1214 return dlsym(dl_prog_handle, lbl);
1215 # elif defined(OBJFORMAT_MACHO)
1217 /* On OS X 10.3 and later, we use dlsym instead of the old legacy
1220 HACK: On OS X, global symbols are prefixed with an underscore.
1221 However, dlsym wants us to omit the leading underscore from the
1222 symbol name. For now, we simply strip it off here (and ONLY
1225 ASSERT(lbl[0] == '_');
1226 return dlsym(dl_prog_handle, lbl+1);
1228 if(NSIsSymbolNameDefined(lbl)) {
1229 NSSymbol symbol = NSLookupAndBindSymbol(lbl);
1230 return NSAddressOfSymbol(symbol);
1234 # endif /* HAVE_DLFCN_H */
1235 # elif defined(OBJFORMAT_PEi386)
1238 sym = lookupSymbolInDLLs(lbl);
1239 if (sym != NULL) { return sym; };
1241 // Also try looking up the symbol without the @N suffix. Some
1242 // DLLs have the suffixes on their symbols, some don't.
1243 zapTrailingAtSign ( lbl );
1244 sym = lookupSymbolInDLLs(lbl);
1245 if (sym != NULL) { return sym; };
1257 /* -----------------------------------------------------------------------------
1258 * Debugging aid: look in GHCi's object symbol tables for symbols
1259 * within DELTA bytes of the specified address, and show their names.
1262 void ghci_enquire ( char* addr );
1264 void ghci_enquire ( char* addr )
1269 const int DELTA = 64;
1274 for (oc = objects; oc; oc = oc->next) {
1275 for (i = 0; i < oc->n_symbols; i++) {
1276 sym = oc->symbols[i];
1277 if (sym == NULL) continue;
1280 a = lookupStrHashTable(symhash, sym);
1283 // debugBelch("ghci_enquire: can't find %s\n", sym);
1285 else if (addr-DELTA <= a && a <= addr+DELTA) {
1286 debugBelch("%p + %3d == `%s'\n", addr, (int)(a - addr), sym);
1293 #ifdef ia64_HOST_ARCH
1294 static unsigned int PLTSize(void);
1298 #define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))
1301 mmapForLinker (size_t bytes, nat flags, int fd)
1303 void *map_addr = NULL;
1306 static nat fixed = 0;
1308 pagesize = getpagesize();
1309 size = ROUND_UP(bytes, pagesize);
1311 #if defined(x86_64_HOST_ARCH)
1314 if (mmap_32bit_base != 0) {
1315 map_addr = mmap_32bit_base;
1319 result = mmap(map_addr, size, PROT_EXEC|PROT_READ|PROT_WRITE,
1320 MAP_PRIVATE|TRY_MAP_32BIT|fixed|flags, fd, 0);
1322 if (result == MAP_FAILED) {
1323 sysErrorBelch("mmap %lu bytes at %p",(lnat)size,map_addr);
1324 errorBelch("Try specifying an address with +RTS -xm<addr> -RTS");
1325 stg_exit(EXIT_FAILURE);
1328 #if defined(x86_64_HOST_ARCH)
1329 if (mmap_32bit_base != 0) {
1330 if (result == map_addr) {
1331 mmap_32bit_base = map_addr + size;
1333 if ((W_)result > 0x80000000) {
1334 // oops, we were given memory over 2Gb
1335 #if defined(freebsd_HOST_OS)
1336 // Some platforms require MAP_FIXED. This is normally
1337 // a bad idea, because MAP_FIXED will overwrite
1338 // existing mappings.
1339 munmap(result,size);
1343 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);
1346 // hmm, we were given memory somewhere else, but it's
1347 // still under 2Gb so we can use it. Next time, ask
1348 // for memory right after the place we just got some
1349 mmap_32bit_base = (void*)result + size;
1353 if ((W_)result > 0x80000000) {
1354 // oops, we were given memory over 2Gb
1355 // ... try allocating memory somewhere else?;
1356 debugTrace(DEBUG_linker,"MAP_32BIT didn't work; gave us %lu bytes at 0x%p", bytes, result);
1357 munmap(result, size);
1359 // Set a base address and try again... (guess: 1Gb)
1360 mmap_32bit_base = (void*)0x40000000;
1370 /* -----------------------------------------------------------------------------
1371 * Load an obj (populate the global symbol table, but don't resolve yet)
1373 * Returns: 1 if ok, 0 on error.
1376 loadObj( char *path )
1388 /* debugBelch("loadObj %s\n", path ); */
1390 /* Check that we haven't already loaded this object.
1391 Ignore requests to load multiple times */
1395 for (o = objects; o; o = o->next) {
1396 if (0 == strcmp(o->fileName, path)) {
1398 break; /* don't need to search further */
1402 IF_DEBUG(linker, debugBelch(
1403 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1404 "same object file twice:\n"
1406 "GHCi will ignore this, but be warned.\n"
1408 return 1; /* success */
1412 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
1414 # if defined(OBJFORMAT_ELF)
1415 oc->formatName = "ELF";
1416 # elif defined(OBJFORMAT_PEi386)
1417 oc->formatName = "PEi386";
1418 # elif defined(OBJFORMAT_MACHO)
1419 oc->formatName = "Mach-O";
1422 barf("loadObj: not implemented on this platform");
1425 r = stat(path, &st);
1426 if (r == -1) { return 0; }
1428 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1429 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1430 strcpy(oc->fileName, path);
1432 oc->fileSize = st.st_size;
1434 oc->sections = NULL;
1435 oc->proddables = NULL;
1437 /* chain it onto the list of objects */
1442 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1444 #if defined(openbsd_HOST_OS)
1445 fd = open(path, O_RDONLY, S_IRUSR);
1447 fd = open(path, O_RDONLY);
1450 barf("loadObj: can't open `%s'", path);
1452 #ifdef ia64_HOST_ARCH
1453 /* The PLT needs to be right before the object */
1456 pagesize = getpagesize();
1457 n = ROUND_UP(PLTSize(), pagesize);
1458 oc->plt = mmap(NULL, n, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
1459 if (oc->plt == MAP_FAILED)
1460 barf("loadObj: can't allocate PLT");
1463 map_addr = oc->plt + n;
1465 n = ROUND_UP(oc->fileSize, pagesize);
1466 oc->image = mmap(map_addr, n, PROT_EXEC|PROT_READ|PROT_WRITE,
1467 MAP_PRIVATE|TRY_MAP_32BIT, fd, 0);
1468 if (oc->image == MAP_FAILED)
1469 barf("loadObj: can't map `%s'", path);
1472 oc->image = mmapForLinker(oc->fileSize, 0, fd);
1477 #else /* !USE_MMAP */
1478 /* load the image into memory */
1479 f = fopen(path, "rb");
1481 barf("loadObj: can't read `%s'", path);
1483 # if defined(mingw32_HOST_OS)
1484 // TODO: We would like to use allocateExec here, but allocateExec
1485 // cannot currently allocate blocks large enough.
1486 oc->image = VirtualAlloc(NULL, oc->fileSize, MEM_RESERVE | MEM_COMMIT,
1487 PAGE_EXECUTE_READWRITE);
1488 # elif defined(darwin_HOST_OS)
1489 // In a Mach-O .o file, all sections can and will be misaligned
1490 // if the total size of the headers is not a multiple of the
1491 // desired alignment. This is fine for .o files that only serve
1492 // as input for the static linker, but it's not fine for us,
1493 // as SSE (used by gcc for floating point) and Altivec require
1494 // 16-byte alignment.
1495 // We calculate the correct alignment from the header before
1496 // reading the file, and then we misalign oc->image on purpose so
1497 // that the actual sections end up aligned again.
1498 oc->misalignment = machoGetMisalignment(f);
1499 oc->image = stgMallocBytes(oc->fileSize + oc->misalignment, "loadObj(image)");
1500 oc->image += oc->misalignment;
1502 oc->image = stgMallocBytes(oc->fileSize, "loadObj(image)");
1507 n = fread ( oc->image, 1, oc->fileSize, f );
1508 if (n != oc->fileSize)
1509 barf("loadObj: error whilst reading `%s'", path);
1512 #endif /* USE_MMAP */
1514 # if defined(OBJFORMAT_MACHO) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1515 r = ocAllocateSymbolExtras_MachO ( oc );
1516 if (!r) { return r; }
1517 # elif defined(OBJFORMAT_ELF) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1518 r = ocAllocateSymbolExtras_ELF ( oc );
1519 if (!r) { return r; }
1522 /* verify the in-memory image */
1523 # if defined(OBJFORMAT_ELF)
1524 r = ocVerifyImage_ELF ( oc );
1525 # elif defined(OBJFORMAT_PEi386)
1526 r = ocVerifyImage_PEi386 ( oc );
1527 # elif defined(OBJFORMAT_MACHO)
1528 r = ocVerifyImage_MachO ( oc );
1530 barf("loadObj: no verify method");
1532 if (!r) { return r; }
1534 /* build the symbol list for this image */
1535 # if defined(OBJFORMAT_ELF)
1536 r = ocGetNames_ELF ( oc );
1537 # elif defined(OBJFORMAT_PEi386)
1538 r = ocGetNames_PEi386 ( oc );
1539 # elif defined(OBJFORMAT_MACHO)
1540 r = ocGetNames_MachO ( oc );
1542 barf("loadObj: no getNames method");
1544 if (!r) { return r; }
1546 /* loaded, but not resolved yet */
1547 oc->status = OBJECT_LOADED;
1552 /* -----------------------------------------------------------------------------
1553 * resolve all the currently unlinked objects in memory
1555 * Returns: 1 if ok, 0 on error.
1565 for (oc = objects; oc; oc = oc->next) {
1566 if (oc->status != OBJECT_RESOLVED) {
1567 # if defined(OBJFORMAT_ELF)
1568 r = ocResolve_ELF ( oc );
1569 # elif defined(OBJFORMAT_PEi386)
1570 r = ocResolve_PEi386 ( oc );
1571 # elif defined(OBJFORMAT_MACHO)
1572 r = ocResolve_MachO ( oc );
1574 barf("resolveObjs: not implemented on this platform");
1576 if (!r) { return r; }
1577 oc->status = OBJECT_RESOLVED;
1583 /* -----------------------------------------------------------------------------
1584 * delete an object from the pool
1587 unloadObj( char *path )
1589 ObjectCode *oc, *prev;
1591 ASSERT(symhash != NULL);
1592 ASSERT(objects != NULL);
1597 for (oc = objects; oc; prev = oc, oc = oc->next) {
1598 if (!strcmp(oc->fileName,path)) {
1600 /* Remove all the mappings for the symbols within this
1605 for (i = 0; i < oc->n_symbols; i++) {
1606 if (oc->symbols[i] != NULL) {
1607 removeStrHashTable(symhash, oc->symbols[i], NULL);
1615 prev->next = oc->next;
1618 // We're going to leave this in place, in case there are
1619 // any pointers from the heap into it:
1620 // #ifdef mingw32_HOST_OS
1621 // VirtualFree(oc->image);
1623 // stgFree(oc->image);
1625 stgFree(oc->fileName);
1626 stgFree(oc->symbols);
1627 stgFree(oc->sections);
1633 errorBelch("unloadObj: can't find `%s' to unload", path);
1637 /* -----------------------------------------------------------------------------
1638 * Sanity checking. For each ObjectCode, maintain a list of address ranges
1639 * which may be prodded during relocation, and abort if we try and write
1640 * outside any of these.
1642 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
1645 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
1646 /* debugBelch("aPB %p %p %d\n", oc, start, size); */
1650 pb->next = oc->proddables;
1651 oc->proddables = pb;
1654 static void checkProddableBlock ( ObjectCode* oc, void* addr )
1657 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
1658 char* s = (char*)(pb->start);
1659 char* e = s + pb->size - 1;
1660 char* a = (char*)addr;
1661 /* Assumes that the biggest fixup involves a 4-byte write. This
1662 probably needs to be changed to 8 (ie, +7) on 64-bit
1664 if (a >= s && (a+3) <= e) return;
1666 barf("checkProddableBlock: invalid fixup in runtime linker");
1669 /* -----------------------------------------------------------------------------
1670 * Section management.
1672 static void addSection ( ObjectCode* oc, SectionKind kind,
1673 void* start, void* end )
1675 Section* s = stgMallocBytes(sizeof(Section), "addSection");
1679 s->next = oc->sections;
1682 debugBelch("addSection: %p-%p (size %d), kind %d\n",
1683 start, ((char*)end)-1, end - start + 1, kind );
1688 /* --------------------------------------------------------------------------
1690 * This is about allocating a small chunk of memory for every symbol in the
1691 * object file. We make sure that the SymboLExtras are always "in range" of
1692 * limited-range PC-relative instructions on various platforms by allocating
1693 * them right next to the object code itself.
1696 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
1699 ocAllocateSymbolExtras
1701 Allocate additional space at the end of the object file image to make room
1702 for jump islands (powerpc, x86_64) and GOT entries (x86_64).
1704 PowerPC relative branch instructions have a 24 bit displacement field.
1705 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
1706 If a particular imported symbol is outside this range, we have to redirect
1707 the jump to a short piece of new code that just loads the 32bit absolute
1708 address and jumps there.
1709 On x86_64, PC-relative jumps and PC-relative accesses to the GOT are limited
1712 This function just allocates space for one SymbolExtra for every
1713 undefined symbol in the object file. The code for the jump islands is
1714 filled in by makeSymbolExtra below.
1717 static int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first )
1724 int misalignment = 0;
1725 #ifdef darwin_HOST_OS
1726 misalignment = oc->misalignment;
1732 // round up to the nearest 4
1733 aligned = (oc->fileSize + 3) & ~3;
1736 pagesize = getpagesize();
1737 n = ROUND_UP( oc->fileSize, pagesize );
1738 m = ROUND_UP( aligned + sizeof (SymbolExtra) * count, pagesize );
1740 /* we try to use spare space at the end of the last page of the
1741 * image for the jump islands, but if there isn't enough space
1742 * then we have to map some (anonymously, remembering MAP_32BIT).
1744 if( m > n ) // we need to allocate more pages
1746 oc->symbol_extras = mmapForLinker(sizeof(SymbolExtra) * count,
1751 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1754 oc->image -= misalignment;
1755 oc->image = stgReallocBytes( oc->image,
1757 aligned + sizeof (SymbolExtra) * count,
1758 "ocAllocateSymbolExtras" );
1759 oc->image += misalignment;
1761 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1762 #endif /* USE_MMAP */
1764 memset( oc->symbol_extras, 0, sizeof (SymbolExtra) * count );
1767 oc->symbol_extras = NULL;
1769 oc->first_symbol_extra = first;
1770 oc->n_symbol_extras = count;
1775 static SymbolExtra* makeSymbolExtra( ObjectCode* oc,
1776 unsigned long symbolNumber,
1777 unsigned long target )
1781 ASSERT( symbolNumber >= oc->first_symbol_extra
1782 && symbolNumber - oc->first_symbol_extra < oc->n_symbol_extras);
1784 extra = &oc->symbol_extras[symbolNumber - oc->first_symbol_extra];
1786 #ifdef powerpc_HOST_ARCH
1787 // lis r12, hi16(target)
1788 extra->jumpIsland.lis_r12 = 0x3d80;
1789 extra->jumpIsland.hi_addr = target >> 16;
1791 // ori r12, r12, lo16(target)
1792 extra->jumpIsland.ori_r12_r12 = 0x618c;
1793 extra->jumpIsland.lo_addr = target & 0xffff;
1796 extra->jumpIsland.mtctr_r12 = 0x7d8903a6;
1799 extra->jumpIsland.bctr = 0x4e800420;
1801 #ifdef x86_64_HOST_ARCH
1803 static uint8_t jmp[] = { 0xFF, 0x25, 0xF2, 0xFF, 0xFF, 0xFF };
1804 extra->addr = target;
1805 memcpy(extra->jumpIsland, jmp, 6);
1813 /* --------------------------------------------------------------------------
1814 * PowerPC specifics (instruction cache flushing)
1815 * ------------------------------------------------------------------------*/
1817 #ifdef powerpc_TARGET_ARCH
1819 ocFlushInstructionCache
1821 Flush the data & instruction caches.
1822 Because the PPC has split data/instruction caches, we have to
1823 do that whenever we modify code at runtime.
1826 static void ocFlushInstructionCache( ObjectCode *oc )
1828 int n = (oc->fileSize + sizeof( SymbolExtra ) * oc->n_symbol_extras + 3) / 4;
1829 unsigned long *p = (unsigned long *) oc->image;
1833 __asm__ volatile ( "dcbf 0,%0\n\t"
1841 __asm__ volatile ( "sync\n\t"
1847 /* --------------------------------------------------------------------------
1848 * PEi386 specifics (Win32 targets)
1849 * ------------------------------------------------------------------------*/
1851 /* The information for this linker comes from
1852 Microsoft Portable Executable
1853 and Common Object File Format Specification
1854 revision 5.1 January 1998
1855 which SimonM says comes from the MS Developer Network CDs.
1857 It can be found there (on older CDs), but can also be found
1860 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
1862 (this is Rev 6.0 from February 1999).
1864 Things move, so if that fails, try searching for it via
1866 http://www.google.com/search?q=PE+COFF+specification
1868 The ultimate reference for the PE format is the Winnt.h
1869 header file that comes with the Platform SDKs; as always,
1870 implementations will drift wrt their documentation.
1872 A good background article on the PE format is Matt Pietrek's
1873 March 1994 article in Microsoft System Journal (MSJ)
1874 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
1875 Win32 Portable Executable File Format." The info in there
1876 has recently been updated in a two part article in
1877 MSDN magazine, issues Feb and March 2002,
1878 "Inside Windows: An In-Depth Look into the Win32 Portable
1879 Executable File Format"
1881 John Levine's book "Linkers and Loaders" contains useful
1886 #if defined(OBJFORMAT_PEi386)
1890 typedef unsigned char UChar;
1891 typedef unsigned short UInt16;
1892 typedef unsigned int UInt32;
1899 UInt16 NumberOfSections;
1900 UInt32 TimeDateStamp;
1901 UInt32 PointerToSymbolTable;
1902 UInt32 NumberOfSymbols;
1903 UInt16 SizeOfOptionalHeader;
1904 UInt16 Characteristics;
1908 #define sizeof_COFF_header 20
1915 UInt32 VirtualAddress;
1916 UInt32 SizeOfRawData;
1917 UInt32 PointerToRawData;
1918 UInt32 PointerToRelocations;
1919 UInt32 PointerToLinenumbers;
1920 UInt16 NumberOfRelocations;
1921 UInt16 NumberOfLineNumbers;
1922 UInt32 Characteristics;
1926 #define sizeof_COFF_section 40
1933 UInt16 SectionNumber;
1936 UChar NumberOfAuxSymbols;
1940 #define sizeof_COFF_symbol 18
1945 UInt32 VirtualAddress;
1946 UInt32 SymbolTableIndex;
1951 #define sizeof_COFF_reloc 10
1954 /* From PE spec doc, section 3.3.2 */
1955 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
1956 windows.h -- for the same purpose, but I want to know what I'm
1958 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
1959 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
1960 #define MYIMAGE_FILE_DLL 0x2000
1961 #define MYIMAGE_FILE_SYSTEM 0x1000
1962 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
1963 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
1964 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
1966 /* From PE spec doc, section 5.4.2 and 5.4.4 */
1967 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
1968 #define MYIMAGE_SYM_CLASS_STATIC 3
1969 #define MYIMAGE_SYM_UNDEFINED 0
1971 /* From PE spec doc, section 4.1 */
1972 #define MYIMAGE_SCN_CNT_CODE 0x00000020
1973 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
1974 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
1976 /* From PE spec doc, section 5.2.1 */
1977 #define MYIMAGE_REL_I386_DIR32 0x0006
1978 #define MYIMAGE_REL_I386_REL32 0x0014
1981 /* We use myindex to calculate array addresses, rather than
1982 simply doing the normal subscript thing. That's because
1983 some of the above structs have sizes which are not
1984 a whole number of words. GCC rounds their sizes up to a
1985 whole number of words, which means that the address calcs
1986 arising from using normal C indexing or pointer arithmetic
1987 are just plain wrong. Sigh.
1990 myindex ( int scale, void* base, int index )
1993 ((UChar*)base) + scale * index;
1998 printName ( UChar* name, UChar* strtab )
2000 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2001 UInt32 strtab_offset = * (UInt32*)(name+4);
2002 debugBelch("%s", strtab + strtab_offset );
2005 for (i = 0; i < 8; i++) {
2006 if (name[i] == 0) break;
2007 debugBelch("%c", name[i] );
2014 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
2016 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2017 UInt32 strtab_offset = * (UInt32*)(name+4);
2018 strncpy ( dst, strtab+strtab_offset, dstSize );
2024 if (name[i] == 0) break;
2034 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
2037 /* If the string is longer than 8 bytes, look in the
2038 string table for it -- this will be correctly zero terminated.
2040 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2041 UInt32 strtab_offset = * (UInt32*)(name+4);
2042 return ((UChar*)strtab) + strtab_offset;
2044 /* Otherwise, if shorter than 8 bytes, return the original,
2045 which by defn is correctly terminated.
2047 if (name[7]==0) return name;
2048 /* The annoying case: 8 bytes. Copy into a temporary
2049 (which is never freed ...)
2051 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
2053 strncpy(newstr,name,8);
2059 /* Just compares the short names (first 8 chars) */
2060 static COFF_section *
2061 findPEi386SectionCalled ( ObjectCode* oc, char* name )
2065 = (COFF_header*)(oc->image);
2066 COFF_section* sectab
2068 ((UChar*)(oc->image))
2069 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2071 for (i = 0; i < hdr->NumberOfSections; i++) {
2074 COFF_section* section_i
2076 myindex ( sizeof_COFF_section, sectab, i );
2077 n1 = (UChar*) &(section_i->Name);
2079 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
2080 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
2081 n1[6]==n2[6] && n1[7]==n2[7])
2090 zapTrailingAtSign ( UChar* sym )
2092 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
2094 if (sym[0] == 0) return;
2096 while (sym[i] != 0) i++;
2099 while (j > 0 && my_isdigit(sym[j])) j--;
2100 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
2105 lookupSymbolInDLLs ( UChar *lbl )
2110 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
2111 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
2113 if (lbl[0] == '_') {
2114 /* HACK: if the name has an initial underscore, try stripping
2115 it off & look that up first. I've yet to verify whether there's
2116 a Rule that governs whether an initial '_' *should always* be
2117 stripped off when mapping from import lib name to the DLL name.
2119 sym = GetProcAddress(o_dll->instance, (lbl+1));
2121 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
2125 sym = GetProcAddress(o_dll->instance, lbl);
2127 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
2136 ocVerifyImage_PEi386 ( ObjectCode* oc )
2141 COFF_section* sectab;
2142 COFF_symbol* symtab;
2144 /* debugBelch("\nLOADING %s\n", oc->fileName); */
2145 hdr = (COFF_header*)(oc->image);
2146 sectab = (COFF_section*) (
2147 ((UChar*)(oc->image))
2148 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2150 symtab = (COFF_symbol*) (
2151 ((UChar*)(oc->image))
2152 + hdr->PointerToSymbolTable
2154 strtab = ((UChar*)symtab)
2155 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2157 if (hdr->Machine != 0x14c) {
2158 errorBelch("%s: Not x86 PEi386", oc->fileName);
2161 if (hdr->SizeOfOptionalHeader != 0) {
2162 errorBelch("%s: PEi386 with nonempty optional header", oc->fileName);
2165 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
2166 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
2167 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
2168 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
2169 errorBelch("%s: Not a PEi386 object file", oc->fileName);
2172 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
2173 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
2174 errorBelch("%s: Invalid PEi386 word size or endiannness: %d",
2176 (int)(hdr->Characteristics));
2179 /* If the string table size is way crazy, this might indicate that
2180 there are more than 64k relocations, despite claims to the
2181 contrary. Hence this test. */
2182 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
2184 if ( (*(UInt32*)strtab) > 600000 ) {
2185 /* Note that 600k has no special significance other than being
2186 big enough to handle the almost-2MB-sized lumps that
2187 constitute HSwin32*.o. */
2188 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
2193 /* No further verification after this point; only debug printing. */
2195 IF_DEBUG(linker, i=1);
2196 if (i == 0) return 1;
2198 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
2199 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
2200 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
2203 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
2204 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
2205 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
2206 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
2207 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
2208 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
2209 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
2211 /* Print the section table. */
2213 for (i = 0; i < hdr->NumberOfSections; i++) {
2215 COFF_section* sectab_i
2217 myindex ( sizeof_COFF_section, sectab, i );
2224 printName ( sectab_i->Name, strtab );
2234 sectab_i->VirtualSize,
2235 sectab_i->VirtualAddress,
2236 sectab_i->SizeOfRawData,
2237 sectab_i->PointerToRawData,
2238 sectab_i->NumberOfRelocations,
2239 sectab_i->PointerToRelocations,
2240 sectab_i->PointerToRawData
2242 reltab = (COFF_reloc*) (
2243 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2246 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2247 /* If the relocation field (a short) has overflowed, the
2248 * real count can be found in the first reloc entry.
2250 * See Section 4.1 (last para) of the PE spec (rev6.0).
2252 COFF_reloc* rel = (COFF_reloc*)
2253 myindex ( sizeof_COFF_reloc, reltab, 0 );
2254 noRelocs = rel->VirtualAddress;
2257 noRelocs = sectab_i->NumberOfRelocations;
2261 for (; j < noRelocs; j++) {
2263 COFF_reloc* rel = (COFF_reloc*)
2264 myindex ( sizeof_COFF_reloc, reltab, j );
2266 " type 0x%-4x vaddr 0x%-8x name `",
2268 rel->VirtualAddress );
2269 sym = (COFF_symbol*)
2270 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
2271 /* Hmm..mysterious looking offset - what's it for? SOF */
2272 printName ( sym->Name, strtab -10 );
2279 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
2280 debugBelch("---START of string table---\n");
2281 for (i = 4; i < *(Int32*)strtab; i++) {
2283 debugBelch("\n"); else
2284 debugBelch("%c", strtab[i] );
2286 debugBelch("--- END of string table---\n");
2291 COFF_symbol* symtab_i;
2292 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2293 symtab_i = (COFF_symbol*)
2294 myindex ( sizeof_COFF_symbol, symtab, i );
2300 printName ( symtab_i->Name, strtab );
2309 (Int32)(symtab_i->SectionNumber),
2310 (UInt32)symtab_i->Type,
2311 (UInt32)symtab_i->StorageClass,
2312 (UInt32)symtab_i->NumberOfAuxSymbols
2314 i += symtab_i->NumberOfAuxSymbols;
2324 ocGetNames_PEi386 ( ObjectCode* oc )
2327 COFF_section* sectab;
2328 COFF_symbol* symtab;
2335 hdr = (COFF_header*)(oc->image);
2336 sectab = (COFF_section*) (
2337 ((UChar*)(oc->image))
2338 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2340 symtab = (COFF_symbol*) (
2341 ((UChar*)(oc->image))
2342 + hdr->PointerToSymbolTable
2344 strtab = ((UChar*)(oc->image))
2345 + hdr->PointerToSymbolTable
2346 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2348 /* Allocate space for any (local, anonymous) .bss sections. */
2350 for (i = 0; i < hdr->NumberOfSections; i++) {
2353 COFF_section* sectab_i
2355 myindex ( sizeof_COFF_section, sectab, i );
2356 if (0 != strcmp(sectab_i->Name, ".bss")) continue;
2357 /* sof 10/05: the PE spec text isn't too clear regarding what
2358 * the SizeOfRawData field is supposed to hold for object
2359 * file sections containing just uninitialized data -- for executables,
2360 * it is supposed to be zero; unclear what it's supposed to be
2361 * for object files. However, VirtualSize is guaranteed to be
2362 * zero for object files, which definitely suggests that SizeOfRawData
2363 * will be non-zero (where else would the size of this .bss section be
2364 * stored?) Looking at the COFF_section info for incoming object files,
2365 * this certainly appears to be the case.
2367 * => I suspect we've been incorrectly handling .bss sections in (relocatable)
2368 * object files up until now. This turned out to bite us with ghc-6.4.1's use
2369 * of gcc-3.4.x, which has started to emit initially-zeroed-out local 'static'
2370 * variable decls into to the .bss section. (The specific function in Q which
2371 * triggered this is libraries/base/cbits/dirUtils.c:__hscore_getFolderPath())
2373 if (sectab_i->VirtualSize == 0 && sectab_i->SizeOfRawData == 0) continue;
2374 /* This is a non-empty .bss section. Allocate zeroed space for
2375 it, and set its PointerToRawData field such that oc->image +
2376 PointerToRawData == addr_of_zeroed_space. */
2377 bss_sz = sectab_i->VirtualSize;
2378 if ( bss_sz < sectab_i->SizeOfRawData) { bss_sz = sectab_i->SizeOfRawData; }
2379 zspace = stgCallocBytes(1, bss_sz, "ocGetNames_PEi386(anonymous bss)");
2380 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
2381 addProddableBlock(oc, zspace, bss_sz);
2382 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
2385 /* Copy section information into the ObjectCode. */
2387 for (i = 0; i < hdr->NumberOfSections; i++) {
2393 = SECTIONKIND_OTHER;
2394 COFF_section* sectab_i
2396 myindex ( sizeof_COFF_section, sectab, i );
2397 IF_DEBUG(linker, debugBelch("section name = %s\n", sectab_i->Name ));
2400 /* I'm sure this is the Right Way to do it. However, the
2401 alternative of testing the sectab_i->Name field seems to
2402 work ok with Cygwin.
2404 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
2405 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
2406 kind = SECTIONKIND_CODE_OR_RODATA;
2409 if (0==strcmp(".text",sectab_i->Name) ||
2410 0==strcmp(".rdata",sectab_i->Name)||
2411 0==strcmp(".rodata",sectab_i->Name))
2412 kind = SECTIONKIND_CODE_OR_RODATA;
2413 if (0==strcmp(".data",sectab_i->Name) ||
2414 0==strcmp(".bss",sectab_i->Name))
2415 kind = SECTIONKIND_RWDATA;
2417 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
2418 sz = sectab_i->SizeOfRawData;
2419 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
2421 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
2422 end = start + sz - 1;
2424 if (kind == SECTIONKIND_OTHER
2425 /* Ignore sections called which contain stabs debugging
2427 && 0 != strcmp(".stab", sectab_i->Name)
2428 && 0 != strcmp(".stabstr", sectab_i->Name)
2429 /* ignore constructor section for now */
2430 && 0 != strcmp(".ctors", sectab_i->Name)
2431 /* ignore section generated from .ident */
2432 && 0!= strcmp("/4", sectab_i->Name)
2433 /* ignore unknown section that appeared in gcc 3.4.5(?) */
2434 && 0!= strcmp(".reloc", sectab_i->Name)
2436 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName);
2440 if (kind != SECTIONKIND_OTHER && end >= start) {
2441 addSection(oc, kind, start, end);
2442 addProddableBlock(oc, start, end - start + 1);
2446 /* Copy exported symbols into the ObjectCode. */
2448 oc->n_symbols = hdr->NumberOfSymbols;
2449 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2450 "ocGetNames_PEi386(oc->symbols)");
2451 /* Call me paranoid; I don't care. */
2452 for (i = 0; i < oc->n_symbols; i++)
2453 oc->symbols[i] = NULL;
2457 COFF_symbol* symtab_i;
2458 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2459 symtab_i = (COFF_symbol*)
2460 myindex ( sizeof_COFF_symbol, symtab, i );
2464 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2465 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2466 /* This symbol is global and defined, viz, exported */
2467 /* for MYIMAGE_SYMCLASS_EXTERNAL
2468 && !MYIMAGE_SYM_UNDEFINED,
2469 the address of the symbol is:
2470 address of relevant section + offset in section
2472 COFF_section* sectabent
2473 = (COFF_section*) myindex ( sizeof_COFF_section,
2475 symtab_i->SectionNumber-1 );
2476 addr = ((UChar*)(oc->image))
2477 + (sectabent->PointerToRawData
2481 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2482 && symtab_i->Value > 0) {
2483 /* This symbol isn't in any section at all, ie, global bss.
2484 Allocate zeroed space for it. */
2485 addr = stgCallocBytes(1, symtab_i->Value,
2486 "ocGetNames_PEi386(non-anonymous bss)");
2487 addSection(oc, SECTIONKIND_RWDATA, addr,
2488 ((UChar*)addr) + symtab_i->Value - 1);
2489 addProddableBlock(oc, addr, symtab_i->Value);
2490 /* debugBelch("BSS section at 0x%x\n", addr); */
2493 if (addr != NULL ) {
2494 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2495 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2496 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2497 ASSERT(i >= 0 && i < oc->n_symbols);
2498 /* cstring_from_COFF_symbol_name always succeeds. */
2499 oc->symbols[i] = sname;
2500 ghciInsertStrHashTable(oc->fileName, symhash, sname, addr);
2504 "IGNORING symbol %d\n"
2508 printName ( symtab_i->Name, strtab );
2517 (Int32)(symtab_i->SectionNumber),
2518 (UInt32)symtab_i->Type,
2519 (UInt32)symtab_i->StorageClass,
2520 (UInt32)symtab_i->NumberOfAuxSymbols
2525 i += symtab_i->NumberOfAuxSymbols;
2534 ocResolve_PEi386 ( ObjectCode* oc )
2537 COFF_section* sectab;
2538 COFF_symbol* symtab;
2548 /* ToDo: should be variable-sized? But is at least safe in the
2549 sense of buffer-overrun-proof. */
2551 /* debugBelch("resolving for %s\n", oc->fileName); */
2553 hdr = (COFF_header*)(oc->image);
2554 sectab = (COFF_section*) (
2555 ((UChar*)(oc->image))
2556 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2558 symtab = (COFF_symbol*) (
2559 ((UChar*)(oc->image))
2560 + hdr->PointerToSymbolTable
2562 strtab = ((UChar*)(oc->image))
2563 + hdr->PointerToSymbolTable
2564 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2566 for (i = 0; i < hdr->NumberOfSections; i++) {
2567 COFF_section* sectab_i
2569 myindex ( sizeof_COFF_section, sectab, i );
2572 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2575 /* Ignore sections called which contain stabs debugging
2577 if (0 == strcmp(".stab", sectab_i->Name)
2578 || 0 == strcmp(".stabstr", sectab_i->Name)
2579 || 0 == strcmp(".ctors", sectab_i->Name))
2582 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2583 /* If the relocation field (a short) has overflowed, the
2584 * real count can be found in the first reloc entry.
2586 * See Section 4.1 (last para) of the PE spec (rev6.0).
2588 * Nov2003 update: the GNU linker still doesn't correctly
2589 * handle the generation of relocatable object files with
2590 * overflown relocations. Hence the output to warn of potential
2593 COFF_reloc* rel = (COFF_reloc*)
2594 myindex ( sizeof_COFF_reloc, reltab, 0 );
2595 noRelocs = rel->VirtualAddress;
2597 /* 10/05: we now assume (and check for) a GNU ld that is capable
2598 * of handling object files with (>2^16) of relocs.
2601 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2606 noRelocs = sectab_i->NumberOfRelocations;
2611 for (; j < noRelocs; j++) {
2613 COFF_reloc* reltab_j
2615 myindex ( sizeof_COFF_reloc, reltab, j );
2617 /* the location to patch */
2619 ((UChar*)(oc->image))
2620 + (sectab_i->PointerToRawData
2621 + reltab_j->VirtualAddress
2622 - sectab_i->VirtualAddress )
2624 /* the existing contents of pP */
2626 /* the symbol to connect to */
2627 sym = (COFF_symbol*)
2628 myindex ( sizeof_COFF_symbol,
2629 symtab, reltab_j->SymbolTableIndex );
2632 "reloc sec %2d num %3d: type 0x%-4x "
2633 "vaddr 0x%-8x name `",
2635 (UInt32)reltab_j->Type,
2636 reltab_j->VirtualAddress );
2637 printName ( sym->Name, strtab );
2638 debugBelch("'\n" ));
2640 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2641 COFF_section* section_sym
2642 = findPEi386SectionCalled ( oc, sym->Name );
2644 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2647 S = ((UInt32)(oc->image))
2648 + (section_sym->PointerToRawData
2651 copyName ( sym->Name, strtab, symbol, 1000-1 );
2652 S = (UInt32) lookupSymbol( symbol );
2653 if ((void*)S != NULL) goto foundit;
2654 /* Newline first because the interactive linker has printed "linking..." */
2655 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, symbol);
2659 checkProddableBlock(oc, pP);
2660 switch (reltab_j->Type) {
2661 case MYIMAGE_REL_I386_DIR32:
2664 case MYIMAGE_REL_I386_REL32:
2665 /* Tricky. We have to insert a displacement at
2666 pP which, when added to the PC for the _next_
2667 insn, gives the address of the target (S).
2668 Problem is to know the address of the next insn
2669 when we only know pP. We assume that this
2670 literal field is always the last in the insn,
2671 so that the address of the next insn is pP+4
2672 -- hence the constant 4.
2673 Also I don't know if A should be added, but so
2674 far it has always been zero.
2676 SOF 05/2005: 'A' (old contents of *pP) have been observed
2677 to contain values other than zero (the 'wx' object file
2678 that came with wxhaskell-0.9.4; dunno how it was compiled..).
2679 So, add displacement to old value instead of asserting
2680 A to be zero. Fixes wxhaskell-related crashes, and no other
2681 ill effects have been observed.
2683 Update: the reason why we're seeing these more elaborate
2684 relocations is due to a switch in how the NCG compiles SRTs
2685 and offsets to them from info tables. SRTs live in .(ro)data,
2686 while info tables live in .text, causing GAS to emit REL32/DISP32
2687 relocations with non-zero values. Adding the displacement is
2688 the right thing to do.
2690 *pP = S - ((UInt32)pP) - 4 + A;
2693 debugBelch("%s: unhandled PEi386 relocation type %d",
2694 oc->fileName, reltab_j->Type);
2701 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2705 #endif /* defined(OBJFORMAT_PEi386) */
2708 /* --------------------------------------------------------------------------
2710 * ------------------------------------------------------------------------*/
2712 #if defined(OBJFORMAT_ELF)
2717 #if defined(sparc_HOST_ARCH)
2718 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2719 #elif defined(i386_HOST_ARCH)
2720 # define ELF_TARGET_386 /* Used inside <elf.h> */
2721 #elif defined(x86_64_HOST_ARCH)
2722 # define ELF_TARGET_X64_64
2724 #elif defined (ia64_HOST_ARCH)
2725 # define ELF_TARGET_IA64 /* Used inside <elf.h> */
2727 # define ELF_FUNCTION_DESC /* calling convention uses function descriptors */
2728 # define ELF_NEED_GOT /* needs Global Offset Table */
2729 # define ELF_NEED_PLT /* needs Procedure Linkage Tables */
2732 #if !defined(openbsd_HOST_OS)
2735 /* openbsd elf has things in different places, with diff names */
2736 # include <elf_abi.h>
2737 # include <machine/reloc.h>
2738 # define R_386_32 RELOC_32
2739 # define R_386_PC32 RELOC_PC32
2742 /* If elf.h doesn't define it */
2743 # ifndef R_X86_64_PC64
2744 # define R_X86_64_PC64 24
2748 * Define a set of types which can be used for both ELF32 and ELF64
2752 #define ELFCLASS ELFCLASS64
2753 #define Elf_Addr Elf64_Addr
2754 #define Elf_Word Elf64_Word
2755 #define Elf_Sword Elf64_Sword
2756 #define Elf_Ehdr Elf64_Ehdr
2757 #define Elf_Phdr Elf64_Phdr
2758 #define Elf_Shdr Elf64_Shdr
2759 #define Elf_Sym Elf64_Sym
2760 #define Elf_Rel Elf64_Rel
2761 #define Elf_Rela Elf64_Rela
2762 #define ELF_ST_TYPE ELF64_ST_TYPE
2763 #define ELF_ST_BIND ELF64_ST_BIND
2764 #define ELF_R_TYPE ELF64_R_TYPE
2765 #define ELF_R_SYM ELF64_R_SYM
2767 #define ELFCLASS ELFCLASS32
2768 #define Elf_Addr Elf32_Addr
2769 #define Elf_Word Elf32_Word
2770 #define Elf_Sword Elf32_Sword
2771 #define Elf_Ehdr Elf32_Ehdr
2772 #define Elf_Phdr Elf32_Phdr
2773 #define Elf_Shdr Elf32_Shdr
2774 #define Elf_Sym Elf32_Sym
2775 #define Elf_Rel Elf32_Rel
2776 #define Elf_Rela Elf32_Rela
2778 #define ELF_ST_TYPE ELF32_ST_TYPE
2781 #define ELF_ST_BIND ELF32_ST_BIND
2784 #define ELF_R_TYPE ELF32_R_TYPE
2787 #define ELF_R_SYM ELF32_R_SYM
2793 * Functions to allocate entries in dynamic sections. Currently we simply
2794 * preallocate a large number, and we don't check if a entry for the given
2795 * target already exists (a linear search is too slow). Ideally these
2796 * entries would be associated with symbols.
2799 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
2800 #define GOT_SIZE 0x20000
2801 #define FUNCTION_TABLE_SIZE 0x10000
2802 #define PLT_SIZE 0x08000
2805 static Elf_Addr got[GOT_SIZE];
2806 static unsigned int gotIndex;
2807 static Elf_Addr gp_val = (Elf_Addr)got;
2810 allocateGOTEntry(Elf_Addr target)
2814 if (gotIndex >= GOT_SIZE)
2815 barf("Global offset table overflow");
2817 entry = &got[gotIndex++];
2819 return (Elf_Addr)entry;
2823 #ifdef ELF_FUNCTION_DESC
2829 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
2830 static unsigned int functionTableIndex;
2833 allocateFunctionDesc(Elf_Addr target)
2835 FunctionDesc *entry;
2837 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
2838 barf("Function table overflow");
2840 entry = &functionTable[functionTableIndex++];
2842 entry->gp = (Elf_Addr)gp_val;
2843 return (Elf_Addr)entry;
2847 copyFunctionDesc(Elf_Addr target)
2849 FunctionDesc *olddesc = (FunctionDesc *)target;
2850 FunctionDesc *newdesc;
2852 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
2853 newdesc->gp = olddesc->gp;
2854 return (Elf_Addr)newdesc;
2859 #ifdef ia64_HOST_ARCH
2860 static void ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value);
2861 static void ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc);
2863 static unsigned char plt_code[] =
2865 /* taken from binutils bfd/elfxx-ia64.c */
2866 0x0b, 0x78, 0x00, 0x02, 0x00, 0x24, /* [MMI] addl r15=0,r1;; */
2867 0x00, 0x41, 0x3c, 0x30, 0x28, 0xc0, /* ld8 r16=[r15],8 */
2868 0x01, 0x08, 0x00, 0x84, /* mov r14=r1;; */
2869 0x11, 0x08, 0x00, 0x1e, 0x18, 0x10, /* [MIB] ld8 r1=[r15] */
2870 0x60, 0x80, 0x04, 0x80, 0x03, 0x00, /* mov b6=r16 */
2871 0x60, 0x00, 0x80, 0x00 /* br.few b6;; */
2874 /* If we can't get to the function descriptor via gp, take a local copy of it */
2875 #define PLT_RELOC(code, target) { \
2876 Elf64_Sxword rel_value = target - gp_val; \
2877 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff)) \
2878 ia64_reloc_gprel22((Elf_Addr)code, copyFunctionDesc(target)); \
2880 ia64_reloc_gprel22((Elf_Addr)code, target); \
2885 unsigned char code[sizeof(plt_code)];
2889 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
2891 PLTEntry *plt = (PLTEntry *)oc->plt;
2894 if (oc->pltIndex >= PLT_SIZE)
2895 barf("Procedure table overflow");
2897 entry = &plt[oc->pltIndex++];
2898 memcpy(entry->code, plt_code, sizeof(entry->code));
2899 PLT_RELOC(entry->code, target);
2900 return (Elf_Addr)entry;
2906 return (PLT_SIZE * sizeof(PLTEntry));
2912 * Generic ELF functions
2916 findElfSection ( void* objImage, Elf_Word sh_type )
2918 char* ehdrC = (char*)objImage;
2919 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2920 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
2921 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2925 for (i = 0; i < ehdr->e_shnum; i++) {
2926 if (shdr[i].sh_type == sh_type
2927 /* Ignore the section header's string table. */
2928 && i != ehdr->e_shstrndx
2929 /* Ignore string tables named .stabstr, as they contain
2931 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2933 ptr = ehdrC + shdr[i].sh_offset;
2940 #if defined(ia64_HOST_ARCH)
2942 findElfSegment ( void* objImage, Elf_Addr vaddr )
2944 char* ehdrC = (char*)objImage;
2945 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2946 Elf_Phdr* phdr = (Elf_Phdr*)(ehdrC + ehdr->e_phoff);
2947 Elf_Addr segaddr = 0;
2950 for (i = 0; i < ehdr->e_phnum; i++) {
2951 segaddr = phdr[i].p_vaddr;
2952 if ((vaddr >= segaddr) && (vaddr < segaddr + phdr[i].p_memsz))
2960 ocVerifyImage_ELF ( ObjectCode* oc )
2964 int i, j, nent, nstrtab, nsymtabs;
2968 char* ehdrC = (char*)(oc->image);
2969 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2971 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
2972 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
2973 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
2974 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
2975 errorBelch("%s: not an ELF object", oc->fileName);
2979 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
2980 errorBelch("%s: unsupported ELF format", oc->fileName);
2984 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
2985 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
2987 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
2988 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
2990 errorBelch("%s: unknown endiannness", oc->fileName);
2994 if (ehdr->e_type != ET_REL) {
2995 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
2998 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
3000 IF_DEBUG(linker,debugBelch( "Architecture is " ));
3001 switch (ehdr->e_machine) {
3002 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
3003 #ifdef EM_SPARC32PLUS
3004 case EM_SPARC32PLUS:
3006 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
3008 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
3010 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
3012 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
3013 #elif defined(EM_AMD64)
3014 case EM_AMD64: IF_DEBUG(linker,debugBelch( "amd64" )); break;
3016 default: IF_DEBUG(linker,debugBelch( "unknown" ));
3017 errorBelch("%s: unknown architecture (e_machine == %d)"
3018 , oc->fileName, ehdr->e_machine);
3022 IF_DEBUG(linker,debugBelch(
3023 "\nSection header table: start %ld, n_entries %d, ent_size %d\n",
3024 (long)ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
3026 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
3028 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3030 if (ehdr->e_shstrndx == SHN_UNDEF) {
3031 errorBelch("%s: no section header string table", oc->fileName);
3034 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
3036 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3039 for (i = 0; i < ehdr->e_shnum; i++) {
3040 IF_DEBUG(linker,debugBelch("%2d: ", i ));
3041 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
3042 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
3043 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
3044 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
3045 ehdrC + shdr[i].sh_offset,
3046 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
3048 if (shdr[i].sh_type == SHT_REL) {
3049 IF_DEBUG(linker,debugBelch("Rel " ));
3050 } else if (shdr[i].sh_type == SHT_RELA) {
3051 IF_DEBUG(linker,debugBelch("RelA " ));
3053 IF_DEBUG(linker,debugBelch(" "));
3056 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
3060 IF_DEBUG(linker,debugBelch( "\nString tables" ));
3063 for (i = 0; i < ehdr->e_shnum; i++) {
3064 if (shdr[i].sh_type == SHT_STRTAB
3065 /* Ignore the section header's string table. */
3066 && i != ehdr->e_shstrndx
3067 /* Ignore string tables named .stabstr, as they contain
3069 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3071 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
3072 strtab = ehdrC + shdr[i].sh_offset;
3077 errorBelch("%s: no string tables, or too many", oc->fileName);
3082 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
3083 for (i = 0; i < ehdr->e_shnum; i++) {
3084 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3085 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
3087 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3088 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3089 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%ld rem)\n",
3091 (long)shdr[i].sh_size % sizeof(Elf_Sym)
3093 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
3094 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
3097 for (j = 0; j < nent; j++) {
3098 IF_DEBUG(linker,debugBelch(" %2d ", j ));
3099 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
3100 (int)stab[j].st_shndx,
3101 (int)stab[j].st_size,
3102 (char*)stab[j].st_value ));
3104 IF_DEBUG(linker,debugBelch("type=" ));
3105 switch (ELF_ST_TYPE(stab[j].st_info)) {
3106 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
3107 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
3108 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
3109 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
3110 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
3111 default: IF_DEBUG(linker,debugBelch("? " )); break;
3113 IF_DEBUG(linker,debugBelch(" " ));
3115 IF_DEBUG(linker,debugBelch("bind=" ));
3116 switch (ELF_ST_BIND(stab[j].st_info)) {
3117 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
3118 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
3119 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
3120 default: IF_DEBUG(linker,debugBelch("? " )); break;
3122 IF_DEBUG(linker,debugBelch(" " ));
3124 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
3128 if (nsymtabs == 0) {
3129 errorBelch("%s: didn't find any symbol tables", oc->fileName);
3136 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
3140 if (hdr->sh_type == SHT_PROGBITS
3141 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
3142 /* .text-style section */
3143 return SECTIONKIND_CODE_OR_RODATA;
3146 if (hdr->sh_type == SHT_PROGBITS
3147 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3148 /* .data-style section */
3149 return SECTIONKIND_RWDATA;
3152 if (hdr->sh_type == SHT_PROGBITS
3153 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
3154 /* .rodata-style section */
3155 return SECTIONKIND_CODE_OR_RODATA;
3158 if (hdr->sh_type == SHT_NOBITS
3159 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3160 /* .bss-style section */
3162 return SECTIONKIND_RWDATA;
3165 return SECTIONKIND_OTHER;
3170 ocGetNames_ELF ( ObjectCode* oc )
3175 char* ehdrC = (char*)(oc->image);
3176 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3177 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
3178 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3180 ASSERT(symhash != NULL);
3183 errorBelch("%s: no strtab", oc->fileName);
3188 for (i = 0; i < ehdr->e_shnum; i++) {
3189 /* Figure out what kind of section it is. Logic derived from
3190 Figure 1.14 ("Special Sections") of the ELF document
3191 ("Portable Formats Specification, Version 1.1"). */
3193 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
3195 if (is_bss && shdr[i].sh_size > 0) {
3196 /* This is a non-empty .bss section. Allocate zeroed space for
3197 it, and set its .sh_offset field such that
3198 ehdrC + .sh_offset == addr_of_zeroed_space. */
3199 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
3200 "ocGetNames_ELF(BSS)");
3201 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
3203 debugBelch("BSS section at 0x%x, size %d\n",
3204 zspace, shdr[i].sh_size);
3208 /* fill in the section info */
3209 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
3210 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
3211 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
3212 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
3215 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3217 /* copy stuff into this module's object symbol table */
3218 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3219 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3221 oc->n_symbols = nent;
3222 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3223 "ocGetNames_ELF(oc->symbols)");
3225 for (j = 0; j < nent; j++) {
3227 char isLocal = FALSE; /* avoids uninit-var warning */
3229 char* nm = strtab + stab[j].st_name;
3230 int secno = stab[j].st_shndx;
3232 /* Figure out if we want to add it; if so, set ad to its
3233 address. Otherwise leave ad == NULL. */
3235 if (secno == SHN_COMMON) {
3237 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
3239 debugBelch("COMMON symbol, size %d name %s\n",
3240 stab[j].st_size, nm);
3242 /* Pointless to do addProddableBlock() for this area,
3243 since the linker should never poke around in it. */
3246 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
3247 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
3249 /* and not an undefined symbol */
3250 && stab[j].st_shndx != SHN_UNDEF
3251 /* and not in a "special section" */
3252 && stab[j].st_shndx < SHN_LORESERVE
3254 /* and it's a not a section or string table or anything silly */
3255 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
3256 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
3257 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
3260 /* Section 0 is the undefined section, hence > and not >=. */
3261 ASSERT(secno > 0 && secno < ehdr->e_shnum);
3263 if (shdr[secno].sh_type == SHT_NOBITS) {
3264 debugBelch(" BSS symbol, size %d off %d name %s\n",
3265 stab[j].st_size, stab[j].st_value, nm);
3268 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
3269 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
3272 #ifdef ELF_FUNCTION_DESC
3273 /* dlsym() and the initialisation table both give us function
3274 * descriptors, so to be consistent we store function descriptors
3275 * in the symbol table */
3276 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
3277 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
3279 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s\n",
3280 ad, oc->fileName, nm ));
3285 /* And the decision is ... */
3289 oc->symbols[j] = nm;
3292 /* Ignore entirely. */
3294 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
3298 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
3299 strtab + stab[j].st_name ));
3302 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
3303 (int)ELF_ST_BIND(stab[j].st_info),
3304 (int)ELF_ST_TYPE(stab[j].st_info),
3305 (int)stab[j].st_shndx,
3306 strtab + stab[j].st_name
3309 oc->symbols[j] = NULL;
3318 /* Do ELF relocations which lack an explicit addend. All x86-linux
3319 relocations appear to be of this form. */
3321 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
3322 Elf_Shdr* shdr, int shnum,
3323 Elf_Sym* stab, char* strtab )
3328 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
3329 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
3330 int target_shndx = shdr[shnum].sh_info;
3331 int symtab_shndx = shdr[shnum].sh_link;
3333 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3334 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
3335 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3336 target_shndx, symtab_shndx ));
3338 /* Skip sections that we're not interested in. */
3341 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
3342 if (kind == SECTIONKIND_OTHER) {
3343 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
3348 for (j = 0; j < nent; j++) {
3349 Elf_Addr offset = rtab[j].r_offset;
3350 Elf_Addr info = rtab[j].r_info;
3352 Elf_Addr P = ((Elf_Addr)targ) + offset;
3353 Elf_Word* pP = (Elf_Word*)P;
3358 StgStablePtr stablePtr;
3361 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
3362 j, (void*)offset, (void*)info ));
3364 IF_DEBUG(linker,debugBelch( " ZERO" ));
3367 Elf_Sym sym = stab[ELF_R_SYM(info)];
3368 /* First see if it is a local symbol. */
3369 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3370 /* Yes, so we can get the address directly from the ELF symbol
3372 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3374 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3375 + stab[ELF_R_SYM(info)].st_value);
3378 symbol = strtab + sym.st_name;
3379 stablePtr = (StgStablePtr)lookupHashTable(stablehash, (StgWord)symbol);
3380 if (NULL == stablePtr) {
3381 /* No, so look up the name in our global table. */
3382 S_tmp = lookupSymbol( symbol );
3383 S = (Elf_Addr)S_tmp;
3385 stableVal = deRefStablePtr( stablePtr );
3387 S = (Elf_Addr)S_tmp;
3391 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3394 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
3397 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
3398 (void*)P, (void*)S, (void*)A ));
3399 checkProddableBlock ( oc, pP );
3403 switch (ELF_R_TYPE(info)) {
3404 # ifdef i386_HOST_ARCH
3405 case R_386_32: *pP = value; break;
3406 case R_386_PC32: *pP = value - P; break;
3409 errorBelch("%s: unhandled ELF relocation(Rel) type %lu\n",
3410 oc->fileName, (lnat)ELF_R_TYPE(info));
3418 /* Do ELF relocations for which explicit addends are supplied.
3419 sparc-solaris relocations appear to be of this form. */
3421 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
3422 Elf_Shdr* shdr, int shnum,
3423 Elf_Sym* stab, char* strtab )
3426 char *symbol = NULL;
3428 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
3429 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
3430 int target_shndx = shdr[shnum].sh_info;
3431 int symtab_shndx = shdr[shnum].sh_link;
3433 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3434 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
3435 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3436 target_shndx, symtab_shndx ));
3438 for (j = 0; j < nent; j++) {
3439 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3440 /* This #ifdef only serves to avoid unused-var warnings. */
3441 Elf_Addr offset = rtab[j].r_offset;
3442 Elf_Addr P = targ + offset;
3444 Elf_Addr info = rtab[j].r_info;
3445 Elf_Addr A = rtab[j].r_addend;
3449 # if defined(sparc_HOST_ARCH)
3450 Elf_Word* pP = (Elf_Word*)P;
3452 # elif defined(ia64_HOST_ARCH)
3453 Elf64_Xword *pP = (Elf64_Xword *)P;
3455 # elif defined(powerpc_HOST_ARCH)
3459 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
3460 j, (void*)offset, (void*)info,
3463 IF_DEBUG(linker,debugBelch( " ZERO" ));
3466 Elf_Sym sym = stab[ELF_R_SYM(info)];
3467 /* First see if it is a local symbol. */
3468 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3469 /* Yes, so we can get the address directly from the ELF symbol
3471 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3473 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3474 + stab[ELF_R_SYM(info)].st_value);
3475 #ifdef ELF_FUNCTION_DESC
3476 /* Make a function descriptor for this function */
3477 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
3478 S = allocateFunctionDesc(S + A);
3483 /* No, so look up the name in our global table. */
3484 symbol = strtab + sym.st_name;
3485 S_tmp = lookupSymbol( symbol );
3486 S = (Elf_Addr)S_tmp;
3488 #ifdef ELF_FUNCTION_DESC
3489 /* If a function, already a function descriptor - we would
3490 have to copy it to add an offset. */
3491 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3492 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3496 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3499 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3502 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3503 (void*)P, (void*)S, (void*)A ));
3504 /* checkProddableBlock ( oc, (void*)P ); */
3508 switch (ELF_R_TYPE(info)) {
3509 # if defined(sparc_HOST_ARCH)
3510 case R_SPARC_WDISP30:
3511 w1 = *pP & 0xC0000000;
3512 w2 = (Elf_Word)((value - P) >> 2);
3513 ASSERT((w2 & 0xC0000000) == 0);
3518 w1 = *pP & 0xFFC00000;
3519 w2 = (Elf_Word)(value >> 10);
3520 ASSERT((w2 & 0xFFC00000) == 0);
3526 w2 = (Elf_Word)(value & 0x3FF);
3527 ASSERT((w2 & ~0x3FF) == 0);
3531 /* According to the Sun documentation:
3533 This relocation type resembles R_SPARC_32, except it refers to an
3534 unaligned word. That is, the word to be relocated must be treated
3535 as four separate bytes with arbitrary alignment, not as a word
3536 aligned according to the architecture requirements.
3538 (JRS: which means that freeloading on the R_SPARC_32 case
3539 is probably wrong, but hey ...)
3543 w2 = (Elf_Word)value;
3546 # elif defined(ia64_HOST_ARCH)
3547 case R_IA64_DIR64LSB:
3548 case R_IA64_FPTR64LSB:
3551 case R_IA64_PCREL64LSB:
3554 case R_IA64_SEGREL64LSB:
3555 addr = findElfSegment(ehdrC, value);
3558 case R_IA64_GPREL22:
3559 ia64_reloc_gprel22(P, value);
3561 case R_IA64_LTOFF22:
3562 case R_IA64_LTOFF22X:
3563 case R_IA64_LTOFF_FPTR22:
3564 addr = allocateGOTEntry(value);
3565 ia64_reloc_gprel22(P, addr);
3567 case R_IA64_PCREL21B:
3568 ia64_reloc_pcrel21(P, S, oc);
3571 /* This goes with R_IA64_LTOFF22X and points to the load to
3572 * convert into a move. We don't implement relaxation. */
3574 # elif defined(powerpc_HOST_ARCH)
3575 case R_PPC_ADDR16_LO:
3576 *(Elf32_Half*) P = value;
3579 case R_PPC_ADDR16_HI:
3580 *(Elf32_Half*) P = value >> 16;
3583 case R_PPC_ADDR16_HA:
3584 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3588 *(Elf32_Word *) P = value;
3592 *(Elf32_Word *) P = value - P;
3598 if( delta << 6 >> 6 != delta )
3600 value = (Elf_Addr) (&makeSymbolExtra( oc, ELF_R_SYM(info), value )
3604 if( value == 0 || delta << 6 >> 6 != delta )
3606 barf( "Unable to make SymbolExtra for #%d",
3612 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3613 | (delta & 0x3fffffc);
3617 #if x86_64_HOST_ARCH
3619 *(Elf64_Xword *)P = value;
3624 StgInt64 off = value - P;
3625 if (off >= 0x7fffffffL || off < -0x80000000L) {
3626 #if X86_64_ELF_NONPIC_HACK
3627 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3629 off = pltAddress + A - P;
3631 barf("R_X86_64_PC32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3632 symbol, off, oc->fileName );
3635 *(Elf64_Word *)P = (Elf64_Word)off;
3641 StgInt64 off = value - P;
3642 *(Elf64_Word *)P = (Elf64_Word)off;
3647 if (value >= 0x7fffffffL) {
3648 #if X86_64_ELF_NONPIC_HACK
3649 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3651 value = pltAddress + A;
3653 barf("R_X86_64_32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3654 symbol, value, oc->fileName );
3657 *(Elf64_Word *)P = (Elf64_Word)value;
3661 if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) {
3662 #if X86_64_ELF_NONPIC_HACK
3663 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3665 value = pltAddress + A;
3667 barf("R_X86_64_32S relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3668 symbol, value, oc->fileName );
3671 *(Elf64_Sword *)P = (Elf64_Sword)value;
3674 case R_X86_64_GOTPCREL:
3676 StgInt64 gotAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)->addr;
3677 StgInt64 off = gotAddress + A - P;
3678 *(Elf64_Word *)P = (Elf64_Word)off;
3682 case R_X86_64_PLT32:
3684 StgInt64 off = value - P;
3685 if (off >= 0x7fffffffL || off < -0x80000000L) {
3686 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3688 off = pltAddress + A - P;
3690 *(Elf64_Word *)P = (Elf64_Word)off;
3696 errorBelch("%s: unhandled ELF relocation(RelA) type %lu\n",
3697 oc->fileName, (lnat)ELF_R_TYPE(info));
3706 ocResolve_ELF ( ObjectCode* oc )
3710 Elf_Sym* stab = NULL;
3711 char* ehdrC = (char*)(oc->image);
3712 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3713 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3715 /* first find "the" symbol table */
3716 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3718 /* also go find the string table */
3719 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3721 if (stab == NULL || strtab == NULL) {
3722 errorBelch("%s: can't find string or symbol table", oc->fileName);
3726 /* Process the relocation sections. */
3727 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3728 if (shdr[shnum].sh_type == SHT_REL) {
3729 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3730 shnum, stab, strtab );
3734 if (shdr[shnum].sh_type == SHT_RELA) {
3735 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3736 shnum, stab, strtab );
3741 #if defined(powerpc_HOST_ARCH)
3742 ocFlushInstructionCache( oc );
3750 * Instructions are 41 bits long, packed into 128 bit bundles with a 5-bit template
3751 * at the front. The following utility functions pack and unpack instructions, and
3752 * take care of the most common relocations.
3755 #ifdef ia64_HOST_ARCH
3758 ia64_extract_instruction(Elf64_Xword *target)
3761 int slot = (Elf_Addr)target & 3;
3762 target = (Elf_Addr)target & ~3;
3770 return ((w1 >> 5) & 0x1ffffffffff);
3772 return (w1 >> 46) | ((w2 & 0x7fffff) << 18);
3776 barf("ia64_extract_instruction: invalid slot %p", target);
3781 ia64_deposit_instruction(Elf64_Xword *target, Elf64_Xword value)
3783 int slot = (Elf_Addr)target & 3;
3784 target = (Elf_Addr)target & ~3;
3789 *target |= value << 5;
3792 *target |= value << 46;
3793 *(target+1) |= value >> 18;
3796 *(target+1) |= value << 23;
3802 ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value)
3804 Elf64_Xword instruction;
3805 Elf64_Sxword rel_value;
3807 rel_value = value - gp_val;
3808 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff))
3809 barf("GP-relative data out of range (address = 0x%lx, gp = 0x%lx)", value, gp_val);
3811 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3812 instruction |= (((rel_value >> 0) & 0x07f) << 13) /* imm7b */
3813 | (((rel_value >> 7) & 0x1ff) << 27) /* imm9d */
3814 | (((rel_value >> 16) & 0x01f) << 22) /* imm5c */
3815 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3816 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3820 ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc)
3822 Elf64_Xword instruction;
3823 Elf64_Sxword rel_value;
3826 entry = allocatePLTEntry(value, oc);
3828 rel_value = (entry >> 4) - (target >> 4);
3829 if ((rel_value > 0xfffff) || (rel_value < -0xfffff))
3830 barf("PLT entry too far away (entry = 0x%lx, target = 0x%lx)", entry, target);
3832 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3833 instruction |= ((rel_value & 0xfffff) << 13) /* imm20b */
3834 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3835 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3841 * PowerPC & X86_64 ELF specifics
3844 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3846 static int ocAllocateSymbolExtras_ELF( ObjectCode *oc )
3852 ehdr = (Elf_Ehdr *) oc->image;
3853 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3855 for( i = 0; i < ehdr->e_shnum; i++ )
3856 if( shdr[i].sh_type == SHT_SYMTAB )
3859 if( i == ehdr->e_shnum )
3861 errorBelch( "This ELF file contains no symtab" );
3865 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3867 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3868 (int) shdr[i].sh_entsize, (int) sizeof( Elf_Sym ) );
3873 return ocAllocateSymbolExtras( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3876 #endif /* powerpc */
3880 /* --------------------------------------------------------------------------
3882 * ------------------------------------------------------------------------*/
3884 #if defined(OBJFORMAT_MACHO)
3887 Support for MachO linking on Darwin/MacOS X
3888 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3890 I hereby formally apologize for the hackish nature of this code.
3891 Things that need to be done:
3892 *) implement ocVerifyImage_MachO
3893 *) add still more sanity checks.
3896 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
3897 #define mach_header mach_header_64
3898 #define segment_command segment_command_64
3899 #define section section_64
3900 #define nlist nlist_64
3903 #ifdef powerpc_HOST_ARCH
3904 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3906 struct mach_header *header = (struct mach_header *) oc->image;
3907 struct load_command *lc = (struct load_command *) (header + 1);
3910 for( i = 0; i < header->ncmds; i++ )
3912 if( lc->cmd == LC_SYMTAB )
3914 // Find out the first and last undefined external
3915 // symbol, so we don't have to allocate too many
3917 struct symtab_command *symLC = (struct symtab_command *) lc;
3918 unsigned min = symLC->nsyms, max = 0;
3919 struct nlist *nlist =
3920 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
3922 for(i=0;i<symLC->nsyms;i++)
3924 if(nlist[i].n_type & N_STAB)
3926 else if(nlist[i].n_type & N_EXT)
3928 if((nlist[i].n_type & N_TYPE) == N_UNDF
3929 && (nlist[i].n_value == 0))
3939 return ocAllocateSymbolExtras(oc, max - min + 1, min);
3944 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3946 return ocAllocateSymbolExtras(oc,0,0);
3949 #ifdef x86_64_HOST_ARCH
3950 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3952 struct mach_header *header = (struct mach_header *) oc->image;
3953 struct load_command *lc = (struct load_command *) (header + 1);
3956 for( i = 0; i < header->ncmds; i++ )
3958 if( lc->cmd == LC_SYMTAB )
3960 // Just allocate one entry for every symbol
3961 struct symtab_command *symLC = (struct symtab_command *) lc;
3963 return ocAllocateSymbolExtras(oc, symLC->nsyms, 0);
3966 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3968 return ocAllocateSymbolExtras(oc,0,0);
3972 static int ocVerifyImage_MachO(ObjectCode* oc)
3974 char *image = (char*) oc->image;
3975 struct mach_header *header = (struct mach_header*) image;
3977 #if x86_64_TARGET_ARCH || powerpc64_TARGET_ARCH
3978 if(header->magic != MH_MAGIC_64)
3981 if(header->magic != MH_MAGIC)
3984 // FIXME: do some more verifying here
3988 static int resolveImports(
3991 struct symtab_command *symLC,
3992 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
3993 unsigned long *indirectSyms,
3994 struct nlist *nlist)
3997 size_t itemSize = 4;
4000 int isJumpTable = 0;
4001 if(!strcmp(sect->sectname,"__jump_table"))
4005 ASSERT(sect->reserved2 == itemSize);
4009 for(i=0; i*itemSize < sect->size;i++)
4011 // according to otool, reserved1 contains the first index into the indirect symbol table
4012 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
4013 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4016 if((symbol->n_type & N_TYPE) == N_UNDF
4017 && (symbol->n_type & N_EXT) && (symbol->n_value != 0))
4018 addr = (void*) (symbol->n_value);
4020 addr = lookupSymbol(nm);
4023 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
4031 checkProddableBlock(oc,image + sect->offset + i*itemSize);
4032 *(image + sect->offset + i*itemSize) = 0xe9; // jmp
4033 *(unsigned*)(image + sect->offset + i*itemSize + 1)
4034 = (char*)addr - (image + sect->offset + i*itemSize + 5);
4039 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
4040 ((void**)(image + sect->offset))[i] = addr;
4047 static unsigned long relocateAddress(
4050 struct section* sections,
4051 unsigned long address)
4054 for(i = 0; i < nSections; i++)
4056 if(sections[i].addr <= address
4057 && address < sections[i].addr + sections[i].size)
4059 return (unsigned long)oc->image
4060 + sections[i].offset + address - sections[i].addr;
4063 barf("Invalid Mach-O file:"
4064 "Address out of bounds while relocating object file");
4068 static int relocateSection(
4071 struct symtab_command *symLC, struct nlist *nlist,
4072 int nSections, struct section* sections, struct section *sect)
4074 struct relocation_info *relocs;
4077 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
4079 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
4081 else if(!strcmp(sect->sectname,"__la_sym_ptr2"))
4083 else if(!strcmp(sect->sectname,"__la_sym_ptr3"))
4087 relocs = (struct relocation_info*) (image + sect->reloff);
4091 #ifdef x86_64_HOST_ARCH
4092 struct relocation_info *reloc = &relocs[i];
4094 char *thingPtr = image + sect->offset + reloc->r_address;
4098 int type = reloc->r_type;
4100 checkProddableBlock(oc,thingPtr);
4101 switch(reloc->r_length)
4104 thing = *(uint8_t*)thingPtr;
4105 baseValue = (uint64_t)thingPtr + 1;
4108 thing = *(uint16_t*)thingPtr;
4109 baseValue = (uint64_t)thingPtr + 2;
4112 thing = *(uint32_t*)thingPtr;
4113 baseValue = (uint64_t)thingPtr + 4;
4116 thing = *(uint64_t*)thingPtr;
4117 baseValue = (uint64_t)thingPtr + 8;
4120 barf("Unknown size.");
4123 if(type == X86_64_RELOC_GOT
4124 || type == X86_64_RELOC_GOT_LOAD)
4126 ASSERT(reloc->r_extern);
4127 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)->addr;
4129 type = X86_64_RELOC_SIGNED;
4131 else if(reloc->r_extern)
4133 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4134 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4135 if(symbol->n_value == 0)
4136 value = (uint64_t) lookupSymbol(nm);
4138 value = relocateAddress(oc, nSections, sections,
4143 value = sections[reloc->r_symbolnum-1].offset
4144 - sections[reloc->r_symbolnum-1].addr
4148 if(type == X86_64_RELOC_BRANCH)
4150 if((int32_t)(value - baseValue) != (int64_t)(value - baseValue))
4152 ASSERT(reloc->r_extern);
4153 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)
4156 ASSERT((int32_t)(value - baseValue) == (int64_t)(value - baseValue));
4157 type = X86_64_RELOC_SIGNED;
4162 case X86_64_RELOC_UNSIGNED:
4163 ASSERT(!reloc->r_pcrel);
4166 case X86_64_RELOC_SIGNED:
4167 ASSERT(reloc->r_pcrel);
4168 thing += value - baseValue;
4170 case X86_64_RELOC_SUBTRACTOR:
4171 ASSERT(!reloc->r_pcrel);
4175 barf("unkown relocation");
4178 switch(reloc->r_length)
4181 *(uint8_t*)thingPtr = thing;
4184 *(uint16_t*)thingPtr = thing;
4187 *(uint32_t*)thingPtr = thing;
4190 *(uint64_t*)thingPtr = thing;
4194 if(relocs[i].r_address & R_SCATTERED)
4196 struct scattered_relocation_info *scat =
4197 (struct scattered_relocation_info*) &relocs[i];
4201 if(scat->r_length == 2)
4203 unsigned long word = 0;
4204 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
4205 checkProddableBlock(oc,wordPtr);
4207 // Note on relocation types:
4208 // i386 uses the GENERIC_RELOC_* types,
4209 // while ppc uses special PPC_RELOC_* types.
4210 // *_RELOC_VANILLA and *_RELOC_PAIR have the same value
4211 // in both cases, all others are different.
4212 // Therefore, we use GENERIC_RELOC_VANILLA
4213 // and GENERIC_RELOC_PAIR instead of the PPC variants,
4214 // and use #ifdefs for the other types.
4216 // Step 1: Figure out what the relocated value should be
4217 if(scat->r_type == GENERIC_RELOC_VANILLA)
4219 word = *wordPtr + (unsigned long) relocateAddress(
4226 #ifdef powerpc_HOST_ARCH
4227 else if(scat->r_type == PPC_RELOC_SECTDIFF
4228 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
4229 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
4230 || scat->r_type == PPC_RELOC_HA16_SECTDIFF)
4232 else if(scat->r_type == GENERIC_RELOC_SECTDIFF)
4235 struct scattered_relocation_info *pair =
4236 (struct scattered_relocation_info*) &relocs[i+1];
4238 if(!pair->r_scattered || pair->r_type != GENERIC_RELOC_PAIR)
4239 barf("Invalid Mach-O file: "
4240 "RELOC_*_SECTDIFF not followed by RELOC_PAIR");
4242 word = (unsigned long)
4243 (relocateAddress(oc, nSections, sections, scat->r_value)
4244 - relocateAddress(oc, nSections, sections, pair->r_value));
4247 #ifdef powerpc_HOST_ARCH
4248 else if(scat->r_type == PPC_RELOC_HI16
4249 || scat->r_type == PPC_RELOC_LO16
4250 || scat->r_type == PPC_RELOC_HA16
4251 || scat->r_type == PPC_RELOC_LO14)
4252 { // these are generated by label+offset things
4253 struct relocation_info *pair = &relocs[i+1];
4254 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
4255 barf("Invalid Mach-O file: "
4256 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
4258 if(scat->r_type == PPC_RELOC_LO16)
4260 word = ((unsigned short*) wordPtr)[1];
4261 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4263 else if(scat->r_type == PPC_RELOC_LO14)
4265 barf("Unsupported Relocation: PPC_RELOC_LO14");
4266 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
4267 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4269 else if(scat->r_type == PPC_RELOC_HI16)
4271 word = ((unsigned short*) wordPtr)[1] << 16;
4272 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4274 else if(scat->r_type == PPC_RELOC_HA16)
4276 word = ((unsigned short*) wordPtr)[1] << 16;
4277 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4281 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
4288 continue; // ignore the others
4290 #ifdef powerpc_HOST_ARCH
4291 if(scat->r_type == GENERIC_RELOC_VANILLA
4292 || scat->r_type == PPC_RELOC_SECTDIFF)
4294 if(scat->r_type == GENERIC_RELOC_VANILLA
4295 || scat->r_type == GENERIC_RELOC_SECTDIFF)
4300 #ifdef powerpc_HOST_ARCH
4301 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
4303 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4305 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
4307 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4309 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
4311 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4312 + ((word & (1<<15)) ? 1 : 0);
4318 continue; // FIXME: I hope it's OK to ignore all the others.
4322 struct relocation_info *reloc = &relocs[i];
4323 if(reloc->r_pcrel && !reloc->r_extern)
4326 if(reloc->r_length == 2)
4328 unsigned long word = 0;
4329 #ifdef powerpc_HOST_ARCH
4330 unsigned long jumpIsland = 0;
4331 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
4332 // to avoid warning and to catch
4336 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
4337 checkProddableBlock(oc,wordPtr);
4339 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4343 #ifdef powerpc_HOST_ARCH
4344 else if(reloc->r_type == PPC_RELOC_LO16)
4346 word = ((unsigned short*) wordPtr)[1];
4347 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4349 else if(reloc->r_type == PPC_RELOC_HI16)
4351 word = ((unsigned short*) wordPtr)[1] << 16;
4352 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4354 else if(reloc->r_type == PPC_RELOC_HA16)
4356 word = ((unsigned short*) wordPtr)[1] << 16;
4357 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4359 else if(reloc->r_type == PPC_RELOC_BR24)
4362 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
4366 if(!reloc->r_extern)
4369 sections[reloc->r_symbolnum-1].offset
4370 - sections[reloc->r_symbolnum-1].addr
4377 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4378 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4379 void *symbolAddress = lookupSymbol(nm);
4382 errorBelch("\nunknown symbol `%s'", nm);
4388 #ifdef powerpc_HOST_ARCH
4389 // In the .o file, this should be a relative jump to NULL
4390 // and we'll change it to a relative jump to the symbol
4391 ASSERT(word + reloc->r_address == 0);
4392 jumpIsland = (unsigned long)
4393 &makeSymbolExtra(oc,
4395 (unsigned long) symbolAddress)
4399 offsetToJumpIsland = word + jumpIsland
4400 - (((long)image) + sect->offset - sect->addr);
4403 word += (unsigned long) symbolAddress
4404 - (((long)image) + sect->offset - sect->addr);
4408 word += (unsigned long) symbolAddress;
4412 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4417 #ifdef powerpc_HOST_ARCH
4418 else if(reloc->r_type == PPC_RELOC_LO16)
4420 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4423 else if(reloc->r_type == PPC_RELOC_HI16)
4425 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4428 else if(reloc->r_type == PPC_RELOC_HA16)
4430 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4431 + ((word & (1<<15)) ? 1 : 0);
4434 else if(reloc->r_type == PPC_RELOC_BR24)
4436 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4438 // The branch offset is too large.
4439 // Therefore, we try to use a jump island.
4442 barf("unconditional relative branch out of range: "
4443 "no jump island available");
4446 word = offsetToJumpIsland;
4447 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4448 barf("unconditional relative branch out of range: "
4449 "jump island out of range");
4451 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
4456 barf("\nunknown relocation %d",reloc->r_type);
4464 static int ocGetNames_MachO(ObjectCode* oc)
4466 char *image = (char*) oc->image;
4467 struct mach_header *header = (struct mach_header*) image;
4468 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4469 unsigned i,curSymbol = 0;
4470 struct segment_command *segLC = NULL;
4471 struct section *sections;
4472 struct symtab_command *symLC = NULL;
4473 struct nlist *nlist;
4474 unsigned long commonSize = 0;
4475 char *commonStorage = NULL;
4476 unsigned long commonCounter;
4478 for(i=0;i<header->ncmds;i++)
4480 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4481 segLC = (struct segment_command*) lc;
4482 else if(lc->cmd == LC_SYMTAB)
4483 symLC = (struct symtab_command*) lc;
4484 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4487 sections = (struct section*) (segLC+1);
4488 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4492 barf("ocGetNames_MachO: no segment load command");
4494 for(i=0;i<segLC->nsects;i++)
4496 if(sections[i].size == 0)
4499 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
4501 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
4502 "ocGetNames_MachO(common symbols)");
4503 sections[i].offset = zeroFillArea - image;
4506 if(!strcmp(sections[i].sectname,"__text"))
4507 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
4508 (void*) (image + sections[i].offset),
4509 (void*) (image + sections[i].offset + sections[i].size));
4510 else if(!strcmp(sections[i].sectname,"__const"))
4511 addSection(oc, SECTIONKIND_RWDATA,
4512 (void*) (image + sections[i].offset),
4513 (void*) (image + sections[i].offset + sections[i].size));
4514 else if(!strcmp(sections[i].sectname,"__data"))
4515 addSection(oc, SECTIONKIND_RWDATA,
4516 (void*) (image + sections[i].offset),
4517 (void*) (image + sections[i].offset + sections[i].size));
4518 else if(!strcmp(sections[i].sectname,"__bss")
4519 || !strcmp(sections[i].sectname,"__common"))
4520 addSection(oc, SECTIONKIND_RWDATA,
4521 (void*) (image + sections[i].offset),
4522 (void*) (image + sections[i].offset + sections[i].size));
4524 addProddableBlock(oc, (void*) (image + sections[i].offset),
4528 // count external symbols defined here
4532 for(i=0;i<symLC->nsyms;i++)
4534 if(nlist[i].n_type & N_STAB)
4536 else if(nlist[i].n_type & N_EXT)
4538 if((nlist[i].n_type & N_TYPE) == N_UNDF
4539 && (nlist[i].n_value != 0))
4541 commonSize += nlist[i].n_value;
4544 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4549 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
4550 "ocGetNames_MachO(oc->symbols)");
4554 for(i=0;i<symLC->nsyms;i++)
4556 if(nlist[i].n_type & N_STAB)
4558 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4560 if(nlist[i].n_type & N_EXT)
4562 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4563 if((nlist[i].n_desc & N_WEAK_DEF) && lookupSymbol(nm))
4564 ; // weak definition, and we already have a definition
4567 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4569 + sections[nlist[i].n_sect-1].offset
4570 - sections[nlist[i].n_sect-1].addr
4571 + nlist[i].n_value);
4572 oc->symbols[curSymbol++] = nm;
4579 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
4580 commonCounter = (unsigned long)commonStorage;
4583 for(i=0;i<symLC->nsyms;i++)
4585 if((nlist[i].n_type & N_TYPE) == N_UNDF
4586 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
4588 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4589 unsigned long sz = nlist[i].n_value;
4591 nlist[i].n_value = commonCounter;
4593 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4594 (void*)commonCounter);
4595 oc->symbols[curSymbol++] = nm;
4597 commonCounter += sz;
4604 static int ocResolve_MachO(ObjectCode* oc)
4606 char *image = (char*) oc->image;
4607 struct mach_header *header = (struct mach_header*) image;
4608 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4610 struct segment_command *segLC = NULL;
4611 struct section *sections;
4612 struct symtab_command *symLC = NULL;
4613 struct dysymtab_command *dsymLC = NULL;
4614 struct nlist *nlist;
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 else if(lc->cmd == LC_DYSYMTAB)
4623 dsymLC = (struct dysymtab_command*) lc;
4624 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4627 sections = (struct section*) (segLC+1);
4628 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4633 unsigned long *indirectSyms
4634 = (unsigned long*) (image + dsymLC->indirectsymoff);
4636 for(i=0;i<segLC->nsects;i++)
4638 if( !strcmp(sections[i].sectname,"__la_symbol_ptr")
4639 || !strcmp(sections[i].sectname,"__la_sym_ptr2")
4640 || !strcmp(sections[i].sectname,"__la_sym_ptr3"))
4642 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4645 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr")
4646 || !strcmp(sections[i].sectname,"__pointers"))
4648 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4651 else if(!strcmp(sections[i].sectname,"__jump_table"))
4653 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4659 for(i=0;i<segLC->nsects;i++)
4661 if(!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
4665 #if defined (powerpc_HOST_ARCH)
4666 ocFlushInstructionCache( oc );
4672 #ifdef powerpc_HOST_ARCH
4674 * The Mach-O object format uses leading underscores. But not everywhere.
4675 * There is a small number of runtime support functions defined in
4676 * libcc_dynamic.a whose name does not have a leading underscore.
4677 * As a consequence, we can't get their address from C code.
4678 * We have to use inline assembler just to take the address of a function.
4682 static void machoInitSymbolsWithoutUnderscore()
4684 extern void* symbolsWithoutUnderscore[];
4685 void **p = symbolsWithoutUnderscore;
4686 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
4688 #undef SymI_NeedsProto
4689 #define SymI_NeedsProto(x) \
4690 __asm__ volatile(".long " # x);
4692 RTS_MACHO_NOUNDERLINE_SYMBOLS
4694 __asm__ volatile(".text");
4696 #undef SymI_NeedsProto
4697 #define SymI_NeedsProto(x) \
4698 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
4700 RTS_MACHO_NOUNDERLINE_SYMBOLS
4702 #undef SymI_NeedsProto
4707 * Figure out by how much to shift the entire Mach-O file in memory
4708 * when loading so that its single segment ends up 16-byte-aligned
4710 static int machoGetMisalignment( FILE * f )
4712 struct mach_header header;
4715 fread(&header, sizeof(header), 1, f);
4718 #if x86_64_TARGET_ARCH || powerpc64_TARGET_ARCH
4719 if(header.magic != MH_MAGIC_64)
4722 if(header.magic != MH_MAGIC)
4726 misalignment = (header.sizeofcmds + sizeof(header))
4729 return misalignment ? (16 - misalignment) : 0;