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(makeStableNamezh_fast) \
228 SymI_HasProto(finalizzeWeakzh_fast)
230 /* These are not available in GUM!!! -- HWL */
231 #define Maybe_Stable_Names
234 #if !defined (mingw32_HOST_OS)
235 #define RTS_POSIX_ONLY_SYMBOLS \
236 SymI_HasProto(shutdownHaskellAndSignal) \
237 SymI_NeedsProto(lockFile) \
238 SymI_NeedsProto(unlockFile) \
239 SymI_HasProto(signal_handlers) \
240 SymI_HasProto(stg_sig_install) \
241 SymI_NeedsProto(nocldstop)
244 #if defined (cygwin32_HOST_OS)
245 #define RTS_MINGW_ONLY_SYMBOLS /**/
246 /* Don't have the ability to read import libs / archives, so
247 * we have to stupidly list a lot of what libcygwin.a
250 #define RTS_CYGWIN_ONLY_SYMBOLS \
251 SymI_HasProto(regfree) \
252 SymI_HasProto(regexec) \
253 SymI_HasProto(regerror) \
254 SymI_HasProto(regcomp) \
255 SymI_HasProto(__errno) \
256 SymI_HasProto(access) \
257 SymI_HasProto(chmod) \
258 SymI_HasProto(chdir) \
259 SymI_HasProto(close) \
260 SymI_HasProto(creat) \
262 SymI_HasProto(dup2) \
263 SymI_HasProto(fstat) \
264 SymI_HasProto(fcntl) \
265 SymI_HasProto(getcwd) \
266 SymI_HasProto(getenv) \
267 SymI_HasProto(lseek) \
268 SymI_HasProto(open) \
269 SymI_HasProto(fpathconf) \
270 SymI_HasProto(pathconf) \
271 SymI_HasProto(stat) \
273 SymI_HasProto(tanh) \
274 SymI_HasProto(cosh) \
275 SymI_HasProto(sinh) \
276 SymI_HasProto(atan) \
277 SymI_HasProto(acos) \
278 SymI_HasProto(asin) \
284 SymI_HasProto(sqrt) \
285 SymI_HasProto(localtime_r) \
286 SymI_HasProto(gmtime_r) \
287 SymI_HasProto(mktime) \
288 SymI_NeedsProto(_imp___tzname) \
289 SymI_HasProto(gettimeofday) \
290 SymI_HasProto(timezone) \
291 SymI_HasProto(tcgetattr) \
292 SymI_HasProto(tcsetattr) \
293 SymI_HasProto(memcpy) \
294 SymI_HasProto(memmove) \
295 SymI_HasProto(realloc) \
296 SymI_HasProto(malloc) \
297 SymI_HasProto(free) \
298 SymI_HasProto(fork) \
299 SymI_HasProto(lstat) \
300 SymI_HasProto(isatty) \
301 SymI_HasProto(mkdir) \
302 SymI_HasProto(opendir) \
303 SymI_HasProto(readdir) \
304 SymI_HasProto(rewinddir) \
305 SymI_HasProto(closedir) \
306 SymI_HasProto(link) \
307 SymI_HasProto(mkfifo) \
308 SymI_HasProto(pipe) \
309 SymI_HasProto(read) \
310 SymI_HasProto(rename) \
311 SymI_HasProto(rmdir) \
312 SymI_HasProto(select) \
313 SymI_HasProto(system) \
314 SymI_HasProto(write) \
315 SymI_HasProto(strcmp) \
316 SymI_HasProto(strcpy) \
317 SymI_HasProto(strncpy) \
318 SymI_HasProto(strerror) \
319 SymI_HasProto(sigaddset) \
320 SymI_HasProto(sigemptyset) \
321 SymI_HasProto(sigprocmask) \
322 SymI_HasProto(umask) \
323 SymI_HasProto(uname) \
324 SymI_HasProto(unlink) \
325 SymI_HasProto(utime) \
326 SymI_HasProto(waitpid)
328 #elif !defined(mingw32_HOST_OS)
329 #define RTS_MINGW_ONLY_SYMBOLS /**/
330 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
331 #else /* defined(mingw32_HOST_OS) */
332 #define RTS_POSIX_ONLY_SYMBOLS /**/
333 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
335 /* Extra syms gen'ed by mingw-2's gcc-3.2: */
337 #define RTS_MINGW_EXTRA_SYMS \
338 SymI_NeedsProto(_imp____mb_cur_max) \
339 SymI_NeedsProto(_imp___pctype)
341 #define RTS_MINGW_EXTRA_SYMS
344 #if HAVE_GETTIMEOFDAY
345 #define RTS_MINGW_GETTIMEOFDAY_SYM SymI_NeedsProto(gettimeofday)
347 #define RTS_MINGW_GETTIMEOFDAY_SYM /**/
350 /* These are statically linked from the mingw libraries into the ghc
351 executable, so we have to employ this hack. */
352 #define RTS_MINGW_ONLY_SYMBOLS \
353 SymI_HasProto(asyncReadzh_fast) \
354 SymI_HasProto(asyncWritezh_fast) \
355 SymI_HasProto(asyncDoProczh_fast) \
356 SymI_HasProto(memset) \
357 SymI_HasProto(inet_ntoa) \
358 SymI_HasProto(inet_addr) \
359 SymI_HasProto(htonl) \
360 SymI_HasProto(recvfrom) \
361 SymI_HasProto(listen) \
362 SymI_HasProto(bind) \
363 SymI_HasProto(shutdown) \
364 SymI_HasProto(connect) \
365 SymI_HasProto(htons) \
366 SymI_HasProto(ntohs) \
367 SymI_HasProto(getservbyname) \
368 SymI_HasProto(getservbyport) \
369 SymI_HasProto(getprotobynumber) \
370 SymI_HasProto(getprotobyname) \
371 SymI_HasProto(gethostbyname) \
372 SymI_HasProto(gethostbyaddr) \
373 SymI_HasProto(gethostname) \
374 SymI_HasProto(strcpy) \
375 SymI_HasProto(strncpy) \
376 SymI_HasProto(abort) \
377 SymI_NeedsProto(_alloca) \
378 SymI_NeedsProto(isxdigit) \
379 SymI_NeedsProto(isupper) \
380 SymI_NeedsProto(ispunct) \
381 SymI_NeedsProto(islower) \
382 SymI_NeedsProto(isspace) \
383 SymI_NeedsProto(isprint) \
384 SymI_NeedsProto(isdigit) \
385 SymI_NeedsProto(iscntrl) \
386 SymI_NeedsProto(isalpha) \
387 SymI_NeedsProto(isalnum) \
388 SymI_HasProto(strcmp) \
389 SymI_HasProto(memmove) \
390 SymI_HasProto(realloc) \
391 SymI_HasProto(malloc) \
393 SymI_HasProto(tanh) \
394 SymI_HasProto(cosh) \
395 SymI_HasProto(sinh) \
396 SymI_HasProto(atan) \
397 SymI_HasProto(acos) \
398 SymI_HasProto(asin) \
404 SymI_HasProto(sqrt) \
405 SymI_HasProto(powf) \
406 SymI_HasProto(tanhf) \
407 SymI_HasProto(coshf) \
408 SymI_HasProto(sinhf) \
409 SymI_HasProto(atanf) \
410 SymI_HasProto(acosf) \
411 SymI_HasProto(asinf) \
412 SymI_HasProto(tanf) \
413 SymI_HasProto(cosf) \
414 SymI_HasProto(sinf) \
415 SymI_HasProto(expf) \
416 SymI_HasProto(logf) \
417 SymI_HasProto(sqrtf) \
418 SymI_HasProto(memcpy) \
419 SymI_HasProto(rts_InstallConsoleEvent) \
420 SymI_HasProto(rts_ConsoleHandlerDone) \
421 SymI_NeedsProto(mktime) \
422 SymI_NeedsProto(_imp___timezone) \
423 SymI_NeedsProto(_imp___tzname) \
424 SymI_NeedsProto(_imp__tzname) \
425 SymI_NeedsProto(_imp___iob) \
426 SymI_NeedsProto(_imp___osver) \
427 SymI_NeedsProto(localtime) \
428 SymI_NeedsProto(gmtime) \
429 SymI_NeedsProto(opendir) \
430 SymI_NeedsProto(readdir) \
431 SymI_NeedsProto(rewinddir) \
432 RTS_MINGW_EXTRA_SYMS \
433 RTS_MINGW_GETTIMEOFDAY_SYM \
434 SymI_NeedsProto(closedir)
437 #if defined(darwin_TARGET_OS) && HAVE_PRINTF_LDBLSTUB
438 #define RTS_DARWIN_ONLY_SYMBOLS \
439 SymI_NeedsProto(asprintf$LDBLStub) \
440 SymI_NeedsProto(err$LDBLStub) \
441 SymI_NeedsProto(errc$LDBLStub) \
442 SymI_NeedsProto(errx$LDBLStub) \
443 SymI_NeedsProto(fprintf$LDBLStub) \
444 SymI_NeedsProto(fscanf$LDBLStub) \
445 SymI_NeedsProto(fwprintf$LDBLStub) \
446 SymI_NeedsProto(fwscanf$LDBLStub) \
447 SymI_NeedsProto(printf$LDBLStub) \
448 SymI_NeedsProto(scanf$LDBLStub) \
449 SymI_NeedsProto(snprintf$LDBLStub) \
450 SymI_NeedsProto(sprintf$LDBLStub) \
451 SymI_NeedsProto(sscanf$LDBLStub) \
452 SymI_NeedsProto(strtold$LDBLStub) \
453 SymI_NeedsProto(swprintf$LDBLStub) \
454 SymI_NeedsProto(swscanf$LDBLStub) \
455 SymI_NeedsProto(syslog$LDBLStub) \
456 SymI_NeedsProto(vasprintf$LDBLStub) \
457 SymI_NeedsProto(verr$LDBLStub) \
458 SymI_NeedsProto(verrc$LDBLStub) \
459 SymI_NeedsProto(verrx$LDBLStub) \
460 SymI_NeedsProto(vfprintf$LDBLStub) \
461 SymI_NeedsProto(vfscanf$LDBLStub) \
462 SymI_NeedsProto(vfwprintf$LDBLStub) \
463 SymI_NeedsProto(vfwscanf$LDBLStub) \
464 SymI_NeedsProto(vprintf$LDBLStub) \
465 SymI_NeedsProto(vscanf$LDBLStub) \
466 SymI_NeedsProto(vsnprintf$LDBLStub) \
467 SymI_NeedsProto(vsprintf$LDBLStub) \
468 SymI_NeedsProto(vsscanf$LDBLStub) \
469 SymI_NeedsProto(vswprintf$LDBLStub) \
470 SymI_NeedsProto(vswscanf$LDBLStub) \
471 SymI_NeedsProto(vsyslog$LDBLStub) \
472 SymI_NeedsProto(vwarn$LDBLStub) \
473 SymI_NeedsProto(vwarnc$LDBLStub) \
474 SymI_NeedsProto(vwarnx$LDBLStub) \
475 SymI_NeedsProto(vwprintf$LDBLStub) \
476 SymI_NeedsProto(vwscanf$LDBLStub) \
477 SymI_NeedsProto(warn$LDBLStub) \
478 SymI_NeedsProto(warnc$LDBLStub) \
479 SymI_NeedsProto(warnx$LDBLStub) \
480 SymI_NeedsProto(wcstold$LDBLStub) \
481 SymI_NeedsProto(wprintf$LDBLStub) \
482 SymI_NeedsProto(wscanf$LDBLStub)
484 #define RTS_DARWIN_ONLY_SYMBOLS
488 # define MAIN_CAP_SYM SymI_HasProto(MainCapability)
490 # define MAIN_CAP_SYM
493 #if !defined(mingw32_HOST_OS)
494 #define RTS_USER_SIGNALS_SYMBOLS \
495 SymI_HasProto(setIOManagerPipe)
497 #define RTS_USER_SIGNALS_SYMBOLS \
498 SymI_HasProto(sendIOManagerEvent) \
499 SymI_HasProto(readIOManagerEvent) \
500 SymI_HasProto(getIOManagerEvent) \
501 SymI_HasProto(console_handler)
504 #define RTS_LIBFFI_SYMBOLS \
505 SymE_NeedsProto(ffi_prep_cif) \
506 SymE_NeedsProto(ffi_call) \
507 SymE_NeedsProto(ffi_type_void) \
508 SymE_NeedsProto(ffi_type_float) \
509 SymE_NeedsProto(ffi_type_double) \
510 SymE_NeedsProto(ffi_type_sint64) \
511 SymE_NeedsProto(ffi_type_uint64) \
512 SymE_NeedsProto(ffi_type_sint32) \
513 SymE_NeedsProto(ffi_type_uint32) \
514 SymE_NeedsProto(ffi_type_sint16) \
515 SymE_NeedsProto(ffi_type_uint16) \
516 SymE_NeedsProto(ffi_type_sint8) \
517 SymE_NeedsProto(ffi_type_uint8) \
518 SymE_NeedsProto(ffi_type_pointer)
520 #ifdef TABLES_NEXT_TO_CODE
521 #define RTS_RET_SYMBOLS /* nothing */
523 #define RTS_RET_SYMBOLS \
524 SymI_HasProto(stg_enter_ret) \
525 SymI_HasProto(stg_gc_fun_ret) \
526 SymI_HasProto(stg_ap_v_ret) \
527 SymI_HasProto(stg_ap_f_ret) \
528 SymI_HasProto(stg_ap_d_ret) \
529 SymI_HasProto(stg_ap_l_ret) \
530 SymI_HasProto(stg_ap_n_ret) \
531 SymI_HasProto(stg_ap_p_ret) \
532 SymI_HasProto(stg_ap_pv_ret) \
533 SymI_HasProto(stg_ap_pp_ret) \
534 SymI_HasProto(stg_ap_ppv_ret) \
535 SymI_HasProto(stg_ap_ppp_ret) \
536 SymI_HasProto(stg_ap_pppv_ret) \
537 SymI_HasProto(stg_ap_pppp_ret) \
538 SymI_HasProto(stg_ap_ppppp_ret) \
539 SymI_HasProto(stg_ap_pppppp_ret)
542 /* On Windows, we link libgmp.a statically into libHSrts.dll */
543 #ifdef mingw32_HOST_OS
545 SymI_HasProto(__gmpz_cmp) \
546 SymI_HasProto(__gmpz_cmp_si) \
547 SymI_HasProto(__gmpz_cmp_ui) \
548 SymI_HasProto(__gmpz_get_si) \
549 SymI_HasProto(__gmpz_get_ui)
552 SymE_HasProto(__gmpz_cmp) \
553 SymE_HasProto(__gmpz_cmp_si) \
554 SymE_HasProto(__gmpz_cmp_ui) \
555 SymE_HasProto(__gmpz_get_si) \
556 SymE_HasProto(__gmpz_get_ui)
559 #define RTS_SYMBOLS \
561 SymI_HasProto(StgReturn) \
562 SymI_HasProto(stg_enter_info) \
563 SymI_HasProto(stg_gc_void_info) \
564 SymI_HasProto(__stg_gc_enter_1) \
565 SymI_HasProto(stg_gc_noregs) \
566 SymI_HasProto(stg_gc_unpt_r1_info) \
567 SymI_HasProto(stg_gc_unpt_r1) \
568 SymI_HasProto(stg_gc_unbx_r1_info) \
569 SymI_HasProto(stg_gc_unbx_r1) \
570 SymI_HasProto(stg_gc_f1_info) \
571 SymI_HasProto(stg_gc_f1) \
572 SymI_HasProto(stg_gc_d1_info) \
573 SymI_HasProto(stg_gc_d1) \
574 SymI_HasProto(stg_gc_l1_info) \
575 SymI_HasProto(stg_gc_l1) \
576 SymI_HasProto(__stg_gc_fun) \
577 SymI_HasProto(stg_gc_fun_info) \
578 SymI_HasProto(stg_gc_gen) \
579 SymI_HasProto(stg_gc_gen_info) \
580 SymI_HasProto(stg_gc_gen_hp) \
581 SymI_HasProto(stg_gc_ut) \
582 SymI_HasProto(stg_gen_yield) \
583 SymI_HasProto(stg_yield_noregs) \
584 SymI_HasProto(stg_yield_to_interpreter) \
585 SymI_HasProto(stg_gen_block) \
586 SymI_HasProto(stg_block_noregs) \
587 SymI_HasProto(stg_block_1) \
588 SymI_HasProto(stg_block_takemvar) \
589 SymI_HasProto(stg_block_putmvar) \
591 SymI_HasProto(MallocFailHook) \
592 SymI_HasProto(OnExitHook) \
593 SymI_HasProto(OutOfHeapHook) \
594 SymI_HasProto(StackOverflowHook) \
595 SymI_HasProto(__encodeDouble) \
596 SymI_HasProto(__encodeFloat) \
597 SymI_HasProto(addDLL) \
599 SymI_HasProto(__int_encodeDouble) \
600 SymI_HasProto(__word_encodeDouble) \
601 SymI_HasProto(__2Int_encodeDouble) \
602 SymI_HasProto(__int_encodeFloat) \
603 SymI_HasProto(__word_encodeFloat) \
604 SymI_HasProto(andIntegerzh_fast) \
605 SymI_HasProto(atomicallyzh_fast) \
606 SymI_HasProto(barf) \
607 SymI_HasProto(debugBelch) \
608 SymI_HasProto(errorBelch) \
609 SymI_HasProto(asyncExceptionsBlockedzh_fast) \
610 SymI_HasProto(blockAsyncExceptionszh_fast) \
611 SymI_HasProto(catchzh_fast) \
612 SymI_HasProto(catchRetryzh_fast) \
613 SymI_HasProto(catchSTMzh_fast) \
614 SymI_HasProto(checkzh_fast) \
615 SymI_HasProto(closure_flags) \
616 SymI_HasProto(cmp_thread) \
617 SymI_HasProto(cmpIntegerzh_fast) \
618 SymI_HasProto(cmpIntegerIntzh_fast) \
619 SymI_HasProto(complementIntegerzh_fast) \
620 SymI_HasProto(createAdjustor) \
621 SymI_HasProto(decodeDoublezh_fast) \
622 SymI_HasProto(decodeFloatzh_fast) \
623 SymI_HasProto(decodeDoublezu2Intzh_fast) \
624 SymI_HasProto(decodeFloatzuIntzh_fast) \
625 SymI_HasProto(defaultsHook) \
626 SymI_HasProto(delayzh_fast) \
627 SymI_HasProto(deRefWeakzh_fast) \
628 SymI_HasProto(deRefStablePtrzh_fast) \
629 SymI_HasProto(dirty_MUT_VAR) \
630 SymI_HasProto(divExactIntegerzh_fast) \
631 SymI_HasProto(divModIntegerzh_fast) \
632 SymI_HasProto(forkzh_fast) \
633 SymI_HasProto(forkOnzh_fast) \
634 SymI_HasProto(forkProcess) \
635 SymI_HasProto(forkOS_createThread) \
636 SymI_HasProto(freeHaskellFunctionPtr) \
637 SymI_HasProto(freeStablePtr) \
638 SymI_HasProto(getOrSetTypeableStore) \
639 SymI_HasProto(gcdIntegerzh_fast) \
640 SymI_HasProto(gcdIntegerIntzh_fast) \
641 SymI_HasProto(gcdIntzh_fast) \
642 SymI_HasProto(genSymZh) \
643 SymI_HasProto(genericRaise) \
644 SymI_HasProto(getProgArgv) \
645 SymI_HasProto(getFullProgArgv) \
646 SymI_HasProto(getStablePtr) \
647 SymI_HasProto(hs_init) \
648 SymI_HasProto(hs_exit) \
649 SymI_HasProto(hs_set_argv) \
650 SymI_HasProto(hs_add_root) \
651 SymI_HasProto(hs_perform_gc) \
652 SymI_HasProto(hs_free_stable_ptr) \
653 SymI_HasProto(hs_free_fun_ptr) \
654 SymI_HasProto(hs_hpc_rootModule) \
655 SymI_HasProto(initLinker) \
656 SymI_HasProto(unpackClosurezh_fast) \
657 SymI_HasProto(getApStackValzh_fast) \
658 SymI_HasProto(getSparkzh_fast) \
659 SymI_HasProto(int2Integerzh_fast) \
660 SymI_HasProto(integer2Intzh_fast) \
661 SymI_HasProto(integer2Wordzh_fast) \
662 SymI_HasProto(isCurrentThreadBoundzh_fast) \
663 SymI_HasProto(isDoubleDenormalized) \
664 SymI_HasProto(isDoubleInfinite) \
665 SymI_HasProto(isDoubleNaN) \
666 SymI_HasProto(isDoubleNegativeZero) \
667 SymI_HasProto(isEmptyMVarzh_fast) \
668 SymI_HasProto(isFloatDenormalized) \
669 SymI_HasProto(isFloatInfinite) \
670 SymI_HasProto(isFloatNaN) \
671 SymI_HasProto(isFloatNegativeZero) \
672 SymI_HasProto(killThreadzh_fast) \
673 SymI_HasProto(loadObj) \
674 SymI_HasProto(insertStableSymbol) \
675 SymI_HasProto(insertSymbol) \
676 SymI_HasProto(lookupSymbol) \
677 SymI_HasProto(makeStablePtrzh_fast) \
678 SymI_HasProto(minusIntegerzh_fast) \
679 SymI_HasProto(mkApUpd0zh_fast) \
680 SymI_HasProto(myThreadIdzh_fast) \
681 SymI_HasProto(labelThreadzh_fast) \
682 SymI_HasProto(newArrayzh_fast) \
683 SymI_HasProto(newBCOzh_fast) \
684 SymI_HasProto(newByteArrayzh_fast) \
685 SymI_HasProto_redirect(newCAF, newDynCAF) \
686 SymI_HasProto(newMVarzh_fast) \
687 SymI_HasProto(newMutVarzh_fast) \
688 SymI_HasProto(newTVarzh_fast) \
689 SymI_HasProto(noDuplicatezh_fast) \
690 SymI_HasProto(atomicModifyMutVarzh_fast) \
691 SymI_HasProto(newPinnedByteArrayzh_fast) \
692 SymI_HasProto(newSpark) \
693 SymI_HasProto(orIntegerzh_fast) \
694 SymI_HasProto(performGC) \
695 SymI_HasProto(performMajorGC) \
696 SymI_HasProto(plusIntegerzh_fast) \
697 SymI_HasProto(prog_argc) \
698 SymI_HasProto(prog_argv) \
699 SymI_HasProto(putMVarzh_fast) \
700 SymI_HasProto(quotIntegerzh_fast) \
701 SymI_HasProto(quotRemIntegerzh_fast) \
702 SymI_HasProto(raisezh_fast) \
703 SymI_HasProto(raiseIOzh_fast) \
704 SymI_HasProto(readTVarzh_fast) \
705 SymI_HasProto(readTVarIOzh_fast) \
706 SymI_HasProto(remIntegerzh_fast) \
707 SymI_HasProto(resetNonBlockingFd) \
708 SymI_HasProto(resumeThread) \
709 SymI_HasProto(resolveObjs) \
710 SymI_HasProto(retryzh_fast) \
711 SymI_HasProto(rts_apply) \
712 SymI_HasProto(rts_checkSchedStatus) \
713 SymI_HasProto(rts_eval) \
714 SymI_HasProto(rts_evalIO) \
715 SymI_HasProto(rts_evalLazyIO) \
716 SymI_HasProto(rts_evalStableIO) \
717 SymI_HasProto(rts_eval_) \
718 SymI_HasProto(rts_getBool) \
719 SymI_HasProto(rts_getChar) \
720 SymI_HasProto(rts_getDouble) \
721 SymI_HasProto(rts_getFloat) \
722 SymI_HasProto(rts_getInt) \
723 SymI_HasProto(rts_getInt8) \
724 SymI_HasProto(rts_getInt16) \
725 SymI_HasProto(rts_getInt32) \
726 SymI_HasProto(rts_getInt64) \
727 SymI_HasProto(rts_getPtr) \
728 SymI_HasProto(rts_getFunPtr) \
729 SymI_HasProto(rts_getStablePtr) \
730 SymI_HasProto(rts_getThreadId) \
731 SymI_HasProto(rts_getWord) \
732 SymI_HasProto(rts_getWord8) \
733 SymI_HasProto(rts_getWord16) \
734 SymI_HasProto(rts_getWord32) \
735 SymI_HasProto(rts_getWord64) \
736 SymI_HasProto(rts_lock) \
737 SymI_HasProto(rts_mkBool) \
738 SymI_HasProto(rts_mkChar) \
739 SymI_HasProto(rts_mkDouble) \
740 SymI_HasProto(rts_mkFloat) \
741 SymI_HasProto(rts_mkInt) \
742 SymI_HasProto(rts_mkInt8) \
743 SymI_HasProto(rts_mkInt16) \
744 SymI_HasProto(rts_mkInt32) \
745 SymI_HasProto(rts_mkInt64) \
746 SymI_HasProto(rts_mkPtr) \
747 SymI_HasProto(rts_mkFunPtr) \
748 SymI_HasProto(rts_mkStablePtr) \
749 SymI_HasProto(rts_mkString) \
750 SymI_HasProto(rts_mkWord) \
751 SymI_HasProto(rts_mkWord8) \
752 SymI_HasProto(rts_mkWord16) \
753 SymI_HasProto(rts_mkWord32) \
754 SymI_HasProto(rts_mkWord64) \
755 SymI_HasProto(rts_unlock) \
756 SymI_HasProto(rtsSupportsBoundThreads) \
757 SymI_HasProto(__hscore_get_saved_termios) \
758 SymI_HasProto(__hscore_set_saved_termios) \
759 SymI_HasProto(setProgArgv) \
760 SymI_HasProto(startupHaskell) \
761 SymI_HasProto(shutdownHaskell) \
762 SymI_HasProto(shutdownHaskellAndExit) \
763 SymI_HasProto(stable_ptr_table) \
764 SymI_HasProto(stackOverflow) \
765 SymI_HasProto(stg_CAF_BLACKHOLE_info) \
766 SymI_HasProto(__stg_EAGER_BLACKHOLE_info) \
767 SymI_HasProto(awakenBlockedQueue) \
768 SymI_HasProto(startTimer) \
769 SymI_HasProto(stg_CHARLIKE_closure) \
770 SymI_HasProto(stg_MVAR_CLEAN_info) \
771 SymI_HasProto(stg_MVAR_DIRTY_info) \
772 SymI_HasProto(stg_IND_STATIC_info) \
773 SymI_HasProto(stg_INTLIKE_closure) \
774 SymI_HasProto(stg_MUT_ARR_PTRS_DIRTY_info) \
775 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN_info) \
776 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN0_info) \
777 SymI_HasProto(stg_WEAK_info) \
778 SymI_HasProto(stg_ap_v_info) \
779 SymI_HasProto(stg_ap_f_info) \
780 SymI_HasProto(stg_ap_d_info) \
781 SymI_HasProto(stg_ap_l_info) \
782 SymI_HasProto(stg_ap_n_info) \
783 SymI_HasProto(stg_ap_p_info) \
784 SymI_HasProto(stg_ap_pv_info) \
785 SymI_HasProto(stg_ap_pp_info) \
786 SymI_HasProto(stg_ap_ppv_info) \
787 SymI_HasProto(stg_ap_ppp_info) \
788 SymI_HasProto(stg_ap_pppv_info) \
789 SymI_HasProto(stg_ap_pppp_info) \
790 SymI_HasProto(stg_ap_ppppp_info) \
791 SymI_HasProto(stg_ap_pppppp_info) \
792 SymI_HasProto(stg_ap_0_fast) \
793 SymI_HasProto(stg_ap_v_fast) \
794 SymI_HasProto(stg_ap_f_fast) \
795 SymI_HasProto(stg_ap_d_fast) \
796 SymI_HasProto(stg_ap_l_fast) \
797 SymI_HasProto(stg_ap_n_fast) \
798 SymI_HasProto(stg_ap_p_fast) \
799 SymI_HasProto(stg_ap_pv_fast) \
800 SymI_HasProto(stg_ap_pp_fast) \
801 SymI_HasProto(stg_ap_ppv_fast) \
802 SymI_HasProto(stg_ap_ppp_fast) \
803 SymI_HasProto(stg_ap_pppv_fast) \
804 SymI_HasProto(stg_ap_pppp_fast) \
805 SymI_HasProto(stg_ap_ppppp_fast) \
806 SymI_HasProto(stg_ap_pppppp_fast) \
807 SymI_HasProto(stg_ap_1_upd_info) \
808 SymI_HasProto(stg_ap_2_upd_info) \
809 SymI_HasProto(stg_ap_3_upd_info) \
810 SymI_HasProto(stg_ap_4_upd_info) \
811 SymI_HasProto(stg_ap_5_upd_info) \
812 SymI_HasProto(stg_ap_6_upd_info) \
813 SymI_HasProto(stg_ap_7_upd_info) \
814 SymI_HasProto(stg_exit) \
815 SymI_HasProto(stg_sel_0_upd_info) \
816 SymI_HasProto(stg_sel_10_upd_info) \
817 SymI_HasProto(stg_sel_11_upd_info) \
818 SymI_HasProto(stg_sel_12_upd_info) \
819 SymI_HasProto(stg_sel_13_upd_info) \
820 SymI_HasProto(stg_sel_14_upd_info) \
821 SymI_HasProto(stg_sel_15_upd_info) \
822 SymI_HasProto(stg_sel_1_upd_info) \
823 SymI_HasProto(stg_sel_2_upd_info) \
824 SymI_HasProto(stg_sel_3_upd_info) \
825 SymI_HasProto(stg_sel_4_upd_info) \
826 SymI_HasProto(stg_sel_5_upd_info) \
827 SymI_HasProto(stg_sel_6_upd_info) \
828 SymI_HasProto(stg_sel_7_upd_info) \
829 SymI_HasProto(stg_sel_8_upd_info) \
830 SymI_HasProto(stg_sel_9_upd_info) \
831 SymI_HasProto(stg_upd_frame_info) \
832 SymI_HasProto(suspendThread) \
833 SymI_HasProto(takeMVarzh_fast) \
834 SymI_HasProto(threadStatuszh_fast) \
835 SymI_HasProto(timesIntegerzh_fast) \
836 SymI_HasProto(tryPutMVarzh_fast) \
837 SymI_HasProto(tryTakeMVarzh_fast) \
838 SymI_HasProto(unblockAsyncExceptionszh_fast) \
839 SymI_HasProto(unloadObj) \
840 SymI_HasProto(unsafeThawArrayzh_fast) \
841 SymI_HasProto(waitReadzh_fast) \
842 SymI_HasProto(waitWritezh_fast) \
843 SymI_HasProto(word2Integerzh_fast) \
844 SymI_HasProto(writeTVarzh_fast) \
845 SymI_HasProto(xorIntegerzh_fast) \
846 SymI_HasProto(yieldzh_fast) \
847 SymI_NeedsProto(stg_interp_constr_entry) \
848 SymI_HasProto(allocateExec) \
849 SymI_HasProto(freeExec) \
850 SymI_HasProto(getAllocations) \
851 SymI_HasProto(revertCAFs) \
852 SymI_HasProto(RtsFlags) \
853 SymI_NeedsProto(rts_breakpoint_io_action) \
854 SymI_NeedsProto(rts_stop_next_breakpoint) \
855 SymI_NeedsProto(rts_stop_on_exception) \
856 SymI_HasProto(stopTimer) \
857 SymI_HasProto(n_capabilities) \
858 RTS_USER_SIGNALS_SYMBOLS
860 #ifdef SUPPORT_LONG_LONGS
861 #define RTS_LONG_LONG_SYMS \
862 SymI_HasProto(int64ToIntegerzh_fast) \
863 SymI_HasProto(word64ToIntegerzh_fast)
865 #define RTS_LONG_LONG_SYMS /* nothing */
868 // 64-bit support functions in libgcc.a
869 #if defined(__GNUC__) && SIZEOF_VOID_P <= 4
870 #define RTS_LIBGCC_SYMBOLS \
871 SymI_NeedsProto(__divdi3) \
872 SymI_NeedsProto(__udivdi3) \
873 SymI_NeedsProto(__moddi3) \
874 SymI_NeedsProto(__umoddi3) \
875 SymI_NeedsProto(__muldi3) \
876 SymI_NeedsProto(__ashldi3) \
877 SymI_NeedsProto(__ashrdi3) \
878 SymI_NeedsProto(__lshrdi3) \
879 SymI_NeedsProto(__eprintf)
880 #elif defined(ia64_HOST_ARCH)
881 #define RTS_LIBGCC_SYMBOLS \
882 SymI_NeedsProto(__divdi3) \
883 SymI_NeedsProto(__udivdi3) \
884 SymI_NeedsProto(__moddi3) \
885 SymI_NeedsProto(__umoddi3) \
886 SymI_NeedsProto(__divsf3) \
887 SymI_NeedsProto(__divdf3)
889 #define RTS_LIBGCC_SYMBOLS
892 #if defined(darwin_HOST_OS) && defined(powerpc_HOST_ARCH)
893 // Symbols that don't have a leading underscore
894 // on Mac OS X. They have to receive special treatment,
895 // see machoInitSymbolsWithoutUnderscore()
896 #define RTS_MACHO_NOUNDERLINE_SYMBOLS \
897 SymI_NeedsProto(saveFP) \
898 SymI_NeedsProto(restFP)
901 /* entirely bogus claims about types of these symbols */
902 #define SymI_NeedsProto(vvv) extern void vvv(void);
903 #if defined(__PIC__) && defined(mingw32_TARGET_OS)
904 #define SymE_HasProto(vvv) SymE_HasProto(vvv);
905 #define SymE_NeedsProto(vvv) extern void _imp__ ## vvv (void);
907 #define SymE_NeedsProto(vvv) SymI_NeedsProto(vvv);
908 #define SymE_HasProto(vvv) SymI_HasProto(vvv)
910 #define SymI_HasProto(vvv) /**/
911 #define SymI_HasProto_redirect(vvv,xxx) /**/
915 RTS_POSIX_ONLY_SYMBOLS
916 RTS_MINGW_ONLY_SYMBOLS
917 RTS_CYGWIN_ONLY_SYMBOLS
918 RTS_DARWIN_ONLY_SYMBOLS
921 #undef SymI_NeedsProto
923 #undef SymI_HasProto_redirect
925 #undef SymE_NeedsProto
927 #ifdef LEADING_UNDERSCORE
928 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
930 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
933 #define SymI_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
935 #define SymE_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
936 (void*)DLL_IMPORT_DATA_REF(vvv) },
938 #define SymI_NeedsProto(vvv) SymI_HasProto(vvv)
939 #define SymE_NeedsProto(vvv) SymE_HasProto(vvv)
941 // SymI_HasProto_redirect allows us to redirect references to one symbol to
942 // another symbol. See newCAF/newDynCAF for an example.
943 #define SymI_HasProto_redirect(vvv,xxx) \
944 { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
947 static RtsSymbolVal rtsSyms[] = {
951 RTS_POSIX_ONLY_SYMBOLS
952 RTS_MINGW_ONLY_SYMBOLS
953 RTS_CYGWIN_ONLY_SYMBOLS
954 RTS_DARWIN_ONLY_SYMBOLS
957 #if defined(darwin_HOST_OS) && defined(i386_HOST_ARCH)
958 // dyld stub code contains references to this,
959 // but it should never be called because we treat
960 // lazy pointers as nonlazy.
961 { "dyld_stub_binding_helper", (void*)0xDEADBEEF },
963 { 0, 0 } /* sentinel */
968 /* -----------------------------------------------------------------------------
969 * Insert symbols into hash tables, checking for duplicates.
972 static void ghciInsertStrHashTable ( char* obj_name,
978 if (lookupHashTable(table, (StgWord)key) == NULL)
980 insertStrHashTable(table, (StgWord)key, data);
985 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
987 "whilst processing object file\n"
989 "This could be caused by:\n"
990 " * Loading two different object files which export the same symbol\n"
991 " * Specifying the same object file twice on the GHCi command line\n"
992 " * An incorrect `package.conf' entry, causing some object to be\n"
994 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
1001 /* -----------------------------------------------------------------------------
1002 * initialize the object linker
1006 static int linker_init_done = 0 ;
1008 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1009 static void *dl_prog_handle;
1017 /* Make initLinker idempotent, so we can call it
1018 before evey relevant operation; that means we
1019 don't need to initialise the linker separately */
1020 if (linker_init_done == 1) { return; } else {
1021 linker_init_done = 1;
1024 stablehash = allocStrHashTable();
1025 symhash = allocStrHashTable();
1027 /* populate the symbol table with stuff from the RTS */
1028 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
1029 ghciInsertStrHashTable("(GHCi built-in symbols)",
1030 symhash, sym->lbl, sym->addr);
1032 # if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH)
1033 machoInitSymbolsWithoutUnderscore();
1036 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1037 # if defined(RTLD_DEFAULT)
1038 dl_prog_handle = RTLD_DEFAULT;
1040 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
1041 # endif /* RTLD_DEFAULT */
1044 #if defined(x86_64_HOST_ARCH)
1045 if (RtsFlags.MiscFlags.linkerMemBase != 0) {
1046 // User-override for mmap_32bit_base
1047 mmap_32bit_base = (void*)RtsFlags.MiscFlags.linkerMemBase;
1052 /* -----------------------------------------------------------------------------
1053 * Loading DLL or .so dynamic libraries
1054 * -----------------------------------------------------------------------------
1056 * Add a DLL from which symbols may be found. In the ELF case, just
1057 * do RTLD_GLOBAL-style add, so no further messing around needs to
1058 * happen in order that symbols in the loaded .so are findable --
1059 * lookupSymbol() will subsequently see them by dlsym on the program's
1060 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
1062 * In the PEi386 case, open the DLLs and put handles to them in a
1063 * linked list. When looking for a symbol, try all handles in the
1064 * list. This means that we need to load even DLLs that are guaranteed
1065 * to be in the ghc.exe image already, just so we can get a handle
1066 * to give to loadSymbol, so that we can find the symbols. For such
1067 * libraries, the LoadLibrary call should be a no-op except for returning
1072 #if defined(OBJFORMAT_PEi386)
1073 /* A record for storing handles into DLLs. */
1078 struct _OpenedDLL* next;
1083 /* A list thereof. */
1084 static OpenedDLL* opened_dlls = NULL;
1088 addDLL( char *dll_name )
1090 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1091 /* ------------------- ELF DLL loader ------------------- */
1097 // omitted: RTLD_NOW
1098 // see http://www.haskell.org/pipermail/cvs-ghc/2007-September/038570.html
1099 hdl= dlopen(dll_name, RTLD_LAZY | RTLD_GLOBAL);
1102 /* dlopen failed; return a ptr to the error msg. */
1104 if (errmsg == NULL) errmsg = "addDLL: unknown error";
1111 # elif defined(OBJFORMAT_PEi386)
1112 /* ------------------- Win32 DLL loader ------------------- */
1120 /* debugBelch("\naddDLL; dll_name = `%s'\n", dll_name); */
1122 /* See if we've already got it, and ignore if so. */
1123 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
1124 if (0 == strcmp(o_dll->name, dll_name))
1128 /* The file name has no suffix (yet) so that we can try
1129 both foo.dll and foo.drv
1131 The documentation for LoadLibrary says:
1132 If no file name extension is specified in the lpFileName
1133 parameter, the default library extension .dll is
1134 appended. However, the file name string can include a trailing
1135 point character (.) to indicate that the module name has no
1138 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
1139 sprintf(buf, "%s.DLL", dll_name);
1140 instance = LoadLibrary(buf);
1141 if (instance == NULL) {
1142 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1143 // KAA: allow loading of drivers (like winspool.drv)
1144 sprintf(buf, "%s.DRV", dll_name);
1145 instance = LoadLibrary(buf);
1146 if (instance == NULL) {
1147 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1148 // #1883: allow loading of unix-style libfoo.dll DLLs
1149 sprintf(buf, "lib%s.DLL", dll_name);
1150 instance = LoadLibrary(buf);
1151 if (instance == NULL) {
1158 /* Add this DLL to the list of DLLs in which to search for symbols. */
1159 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
1160 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
1161 strcpy(o_dll->name, dll_name);
1162 o_dll->instance = instance;
1163 o_dll->next = opened_dlls;
1164 opened_dlls = o_dll;
1170 sysErrorBelch(dll_name);
1172 /* LoadLibrary failed; return a ptr to the error msg. */
1173 return "addDLL: could not load DLL";
1176 barf("addDLL: not implemented on this platform");
1180 /* -----------------------------------------------------------------------------
1181 * insert a stable symbol in the hash table
1185 insertStableSymbol(char* obj_name, char* key, StgPtr p)
1187 ghciInsertStrHashTable(obj_name, stablehash, key, getStablePtr(p));
1191 /* -----------------------------------------------------------------------------
1192 * insert a symbol in the hash table
1195 insertSymbol(char* obj_name, char* key, void* data)
1197 ghciInsertStrHashTable(obj_name, symhash, key, data);
1200 /* -----------------------------------------------------------------------------
1201 * lookup a symbol in the hash table
1204 lookupSymbol( char *lbl )
1208 ASSERT(symhash != NULL);
1209 val = lookupStrHashTable(symhash, lbl);
1212 # if defined(OBJFORMAT_ELF)
1213 return dlsym(dl_prog_handle, lbl);
1214 # elif defined(OBJFORMAT_MACHO)
1216 /* On OS X 10.3 and later, we use dlsym instead of the old legacy
1219 HACK: On OS X, global symbols are prefixed with an underscore.
1220 However, dlsym wants us to omit the leading underscore from the
1221 symbol name. For now, we simply strip it off here (and ONLY
1224 ASSERT(lbl[0] == '_');
1225 return dlsym(dl_prog_handle, lbl+1);
1227 if(NSIsSymbolNameDefined(lbl)) {
1228 NSSymbol symbol = NSLookupAndBindSymbol(lbl);
1229 return NSAddressOfSymbol(symbol);
1233 # endif /* HAVE_DLFCN_H */
1234 # elif defined(OBJFORMAT_PEi386)
1237 sym = lookupSymbolInDLLs(lbl);
1238 if (sym != NULL) { return sym; };
1240 // Also try looking up the symbol without the @N suffix. Some
1241 // DLLs have the suffixes on their symbols, some don't.
1242 zapTrailingAtSign ( lbl );
1243 sym = lookupSymbolInDLLs(lbl);
1244 if (sym != NULL) { return sym; };
1256 /* -----------------------------------------------------------------------------
1257 * Debugging aid: look in GHCi's object symbol tables for symbols
1258 * within DELTA bytes of the specified address, and show their names.
1261 void ghci_enquire ( char* addr );
1263 void ghci_enquire ( char* addr )
1268 const int DELTA = 64;
1273 for (oc = objects; oc; oc = oc->next) {
1274 for (i = 0; i < oc->n_symbols; i++) {
1275 sym = oc->symbols[i];
1276 if (sym == NULL) continue;
1279 a = lookupStrHashTable(symhash, sym);
1282 // debugBelch("ghci_enquire: can't find %s\n", sym);
1284 else if (addr-DELTA <= a && a <= addr+DELTA) {
1285 debugBelch("%p + %3d == `%s'\n", addr, (int)(a - addr), sym);
1292 #ifdef ia64_HOST_ARCH
1293 static unsigned int PLTSize(void);
1297 #define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))
1300 mmapForLinker (size_t bytes, nat flags, int fd)
1302 void *map_addr = NULL;
1305 static nat fixed = 0;
1307 pagesize = getpagesize();
1308 size = ROUND_UP(bytes, pagesize);
1310 #if defined(x86_64_HOST_ARCH)
1313 if (mmap_32bit_base != 0) {
1314 map_addr = mmap_32bit_base;
1318 result = mmap(map_addr, size, PROT_EXEC|PROT_READ|PROT_WRITE,
1319 MAP_PRIVATE|TRY_MAP_32BIT|fixed|flags, fd, 0);
1321 if (result == MAP_FAILED) {
1322 sysErrorBelch("mmap %lu bytes at %p",(lnat)size,map_addr);
1323 errorBelch("Try specifying an address with +RTS -xm<addr> -RTS");
1324 stg_exit(EXIT_FAILURE);
1327 #if defined(x86_64_HOST_ARCH)
1328 if (mmap_32bit_base != 0) {
1329 if (result == map_addr) {
1330 mmap_32bit_base = map_addr + size;
1332 if ((W_)result > 0x80000000) {
1333 // oops, we were given memory over 2Gb
1334 #if defined(freebsd_HOST_OS)
1335 // Some platforms require MAP_FIXED. This is normally
1336 // a bad idea, because MAP_FIXED will overwrite
1337 // existing mappings.
1338 munmap(result,size);
1342 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);
1345 // hmm, we were given memory somewhere else, but it's
1346 // still under 2Gb so we can use it. Next time, ask
1347 // for memory right after the place we just got some
1348 mmap_32bit_base = (void*)result + size;
1352 if ((W_)result > 0x80000000) {
1353 // oops, we were given memory over 2Gb
1354 // ... try allocating memory somewhere else?;
1355 debugTrace(DEBUG_linker,"MAP_32BIT didn't work; gave us %lu bytes at 0x%p", bytes, result);
1356 munmap(result, size);
1358 // Set a base address and try again... (guess: 1Gb)
1359 mmap_32bit_base = (void*)0x40000000;
1369 /* -----------------------------------------------------------------------------
1370 * Load an obj (populate the global symbol table, but don't resolve yet)
1372 * Returns: 1 if ok, 0 on error.
1375 loadObj( char *path )
1387 /* debugBelch("loadObj %s\n", path ); */
1389 /* Check that we haven't already loaded this object.
1390 Ignore requests to load multiple times */
1394 for (o = objects; o; o = o->next) {
1395 if (0 == strcmp(o->fileName, path)) {
1397 break; /* don't need to search further */
1401 IF_DEBUG(linker, debugBelch(
1402 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1403 "same object file twice:\n"
1405 "GHCi will ignore this, but be warned.\n"
1407 return 1; /* success */
1411 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
1413 # if defined(OBJFORMAT_ELF)
1414 oc->formatName = "ELF";
1415 # elif defined(OBJFORMAT_PEi386)
1416 oc->formatName = "PEi386";
1417 # elif defined(OBJFORMAT_MACHO)
1418 oc->formatName = "Mach-O";
1421 barf("loadObj: not implemented on this platform");
1424 r = stat(path, &st);
1425 if (r == -1) { return 0; }
1427 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1428 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1429 strcpy(oc->fileName, path);
1431 oc->fileSize = st.st_size;
1433 oc->sections = NULL;
1434 oc->proddables = NULL;
1436 /* chain it onto the list of objects */
1441 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1443 #if defined(openbsd_HOST_OS)
1444 fd = open(path, O_RDONLY, S_IRUSR);
1446 fd = open(path, O_RDONLY);
1449 barf("loadObj: can't open `%s'", path);
1451 #ifdef ia64_HOST_ARCH
1452 /* The PLT needs to be right before the object */
1455 pagesize = getpagesize();
1456 n = ROUND_UP(PLTSize(), pagesize);
1457 oc->plt = mmap(NULL, n, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
1458 if (oc->plt == MAP_FAILED)
1459 barf("loadObj: can't allocate PLT");
1462 map_addr = oc->plt + n;
1464 n = ROUND_UP(oc->fileSize, pagesize);
1465 oc->image = mmap(map_addr, n, PROT_EXEC|PROT_READ|PROT_WRITE,
1466 MAP_PRIVATE|TRY_MAP_32BIT, fd, 0);
1467 if (oc->image == MAP_FAILED)
1468 barf("loadObj: can't map `%s'", path);
1471 oc->image = mmapForLinker(oc->fileSize, 0, fd);
1476 #else /* !USE_MMAP */
1477 /* load the image into memory */
1478 f = fopen(path, "rb");
1480 barf("loadObj: can't read `%s'", path);
1482 # if defined(mingw32_HOST_OS)
1483 // TODO: We would like to use allocateExec here, but allocateExec
1484 // cannot currently allocate blocks large enough.
1485 oc->image = VirtualAlloc(NULL, oc->fileSize, MEM_RESERVE | MEM_COMMIT,
1486 PAGE_EXECUTE_READWRITE);
1487 # elif defined(darwin_HOST_OS)
1488 // In a Mach-O .o file, all sections can and will be misaligned
1489 // if the total size of the headers is not a multiple of the
1490 // desired alignment. This is fine for .o files that only serve
1491 // as input for the static linker, but it's not fine for us,
1492 // as SSE (used by gcc for floating point) and Altivec require
1493 // 16-byte alignment.
1494 // We calculate the correct alignment from the header before
1495 // reading the file, and then we misalign oc->image on purpose so
1496 // that the actual sections end up aligned again.
1497 oc->misalignment = machoGetMisalignment(f);
1498 oc->image = stgMallocBytes(oc->fileSize + oc->misalignment, "loadObj(image)");
1499 oc->image += oc->misalignment;
1501 oc->image = stgMallocBytes(oc->fileSize, "loadObj(image)");
1506 n = fread ( oc->image, 1, oc->fileSize, f );
1507 if (n != oc->fileSize)
1508 barf("loadObj: error whilst reading `%s'", path);
1511 #endif /* USE_MMAP */
1513 # if defined(OBJFORMAT_MACHO) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1514 r = ocAllocateSymbolExtras_MachO ( oc );
1515 if (!r) { return r; }
1516 # elif defined(OBJFORMAT_ELF) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1517 r = ocAllocateSymbolExtras_ELF ( oc );
1518 if (!r) { return r; }
1521 /* verify the in-memory image */
1522 # if defined(OBJFORMAT_ELF)
1523 r = ocVerifyImage_ELF ( oc );
1524 # elif defined(OBJFORMAT_PEi386)
1525 r = ocVerifyImage_PEi386 ( oc );
1526 # elif defined(OBJFORMAT_MACHO)
1527 r = ocVerifyImage_MachO ( oc );
1529 barf("loadObj: no verify method");
1531 if (!r) { return r; }
1533 /* build the symbol list for this image */
1534 # if defined(OBJFORMAT_ELF)
1535 r = ocGetNames_ELF ( oc );
1536 # elif defined(OBJFORMAT_PEi386)
1537 r = ocGetNames_PEi386 ( oc );
1538 # elif defined(OBJFORMAT_MACHO)
1539 r = ocGetNames_MachO ( oc );
1541 barf("loadObj: no getNames method");
1543 if (!r) { return r; }
1545 /* loaded, but not resolved yet */
1546 oc->status = OBJECT_LOADED;
1551 /* -----------------------------------------------------------------------------
1552 * resolve all the currently unlinked objects in memory
1554 * Returns: 1 if ok, 0 on error.
1564 for (oc = objects; oc; oc = oc->next) {
1565 if (oc->status != OBJECT_RESOLVED) {
1566 # if defined(OBJFORMAT_ELF)
1567 r = ocResolve_ELF ( oc );
1568 # elif defined(OBJFORMAT_PEi386)
1569 r = ocResolve_PEi386 ( oc );
1570 # elif defined(OBJFORMAT_MACHO)
1571 r = ocResolve_MachO ( oc );
1573 barf("resolveObjs: not implemented on this platform");
1575 if (!r) { return r; }
1576 oc->status = OBJECT_RESOLVED;
1582 /* -----------------------------------------------------------------------------
1583 * delete an object from the pool
1586 unloadObj( char *path )
1588 ObjectCode *oc, *prev;
1590 ASSERT(symhash != NULL);
1591 ASSERT(objects != NULL);
1596 for (oc = objects; oc; prev = oc, oc = oc->next) {
1597 if (!strcmp(oc->fileName,path)) {
1599 /* Remove all the mappings for the symbols within this
1604 for (i = 0; i < oc->n_symbols; i++) {
1605 if (oc->symbols[i] != NULL) {
1606 removeStrHashTable(symhash, oc->symbols[i], NULL);
1614 prev->next = oc->next;
1617 // We're going to leave this in place, in case there are
1618 // any pointers from the heap into it:
1619 // #ifdef mingw32_HOST_OS
1620 // VirtualFree(oc->image);
1622 // stgFree(oc->image);
1624 stgFree(oc->fileName);
1625 stgFree(oc->symbols);
1626 stgFree(oc->sections);
1632 errorBelch("unloadObj: can't find `%s' to unload", path);
1636 /* -----------------------------------------------------------------------------
1637 * Sanity checking. For each ObjectCode, maintain a list of address ranges
1638 * which may be prodded during relocation, and abort if we try and write
1639 * outside any of these.
1641 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
1644 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
1645 /* debugBelch("aPB %p %p %d\n", oc, start, size); */
1649 pb->next = oc->proddables;
1650 oc->proddables = pb;
1653 static void checkProddableBlock ( ObjectCode* oc, void* addr )
1656 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
1657 char* s = (char*)(pb->start);
1658 char* e = s + pb->size - 1;
1659 char* a = (char*)addr;
1660 /* Assumes that the biggest fixup involves a 4-byte write. This
1661 probably needs to be changed to 8 (ie, +7) on 64-bit
1663 if (a >= s && (a+3) <= e) return;
1665 barf("checkProddableBlock: invalid fixup in runtime linker");
1668 /* -----------------------------------------------------------------------------
1669 * Section management.
1671 static void addSection ( ObjectCode* oc, SectionKind kind,
1672 void* start, void* end )
1674 Section* s = stgMallocBytes(sizeof(Section), "addSection");
1678 s->next = oc->sections;
1681 debugBelch("addSection: %p-%p (size %d), kind %d\n",
1682 start, ((char*)end)-1, end - start + 1, kind );
1687 /* --------------------------------------------------------------------------
1689 * This is about allocating a small chunk of memory for every symbol in the
1690 * object file. We make sure that the SymboLExtras are always "in range" of
1691 * limited-range PC-relative instructions on various platforms by allocating
1692 * them right next to the object code itself.
1695 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
1698 ocAllocateSymbolExtras
1700 Allocate additional space at the end of the object file image to make room
1701 for jump islands (powerpc, x86_64) and GOT entries (x86_64).
1703 PowerPC relative branch instructions have a 24 bit displacement field.
1704 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
1705 If a particular imported symbol is outside this range, we have to redirect
1706 the jump to a short piece of new code that just loads the 32bit absolute
1707 address and jumps there.
1708 On x86_64, PC-relative jumps and PC-relative accesses to the GOT are limited
1711 This function just allocates space for one SymbolExtra for every
1712 undefined symbol in the object file. The code for the jump islands is
1713 filled in by makeSymbolExtra below.
1716 static int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first )
1723 int misalignment = 0;
1724 #ifdef darwin_HOST_OS
1725 misalignment = oc->misalignment;
1731 // round up to the nearest 4
1732 aligned = (oc->fileSize + 3) & ~3;
1735 pagesize = getpagesize();
1736 n = ROUND_UP( oc->fileSize, pagesize );
1737 m = ROUND_UP( aligned + sizeof (SymbolExtra) * count, pagesize );
1739 /* we try to use spare space at the end of the last page of the
1740 * image for the jump islands, but if there isn't enough space
1741 * then we have to map some (anonymously, remembering MAP_32BIT).
1743 if( m > n ) // we need to allocate more pages
1745 oc->symbol_extras = mmapForLinker(sizeof(SymbolExtra) * count,
1750 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1753 oc->image -= misalignment;
1754 oc->image = stgReallocBytes( oc->image,
1756 aligned + sizeof (SymbolExtra) * count,
1757 "ocAllocateSymbolExtras" );
1758 oc->image += misalignment;
1760 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1761 #endif /* USE_MMAP */
1763 memset( oc->symbol_extras, 0, sizeof (SymbolExtra) * count );
1766 oc->symbol_extras = NULL;
1768 oc->first_symbol_extra = first;
1769 oc->n_symbol_extras = count;
1774 static SymbolExtra* makeSymbolExtra( ObjectCode* oc,
1775 unsigned long symbolNumber,
1776 unsigned long target )
1780 ASSERT( symbolNumber >= oc->first_symbol_extra
1781 && symbolNumber - oc->first_symbol_extra < oc->n_symbol_extras);
1783 extra = &oc->symbol_extras[symbolNumber - oc->first_symbol_extra];
1785 #ifdef powerpc_HOST_ARCH
1786 // lis r12, hi16(target)
1787 extra->jumpIsland.lis_r12 = 0x3d80;
1788 extra->jumpIsland.hi_addr = target >> 16;
1790 // ori r12, r12, lo16(target)
1791 extra->jumpIsland.ori_r12_r12 = 0x618c;
1792 extra->jumpIsland.lo_addr = target & 0xffff;
1795 extra->jumpIsland.mtctr_r12 = 0x7d8903a6;
1798 extra->jumpIsland.bctr = 0x4e800420;
1800 #ifdef x86_64_HOST_ARCH
1802 static uint8_t jmp[] = { 0xFF, 0x25, 0xF2, 0xFF, 0xFF, 0xFF };
1803 extra->addr = target;
1804 memcpy(extra->jumpIsland, jmp, 6);
1812 /* --------------------------------------------------------------------------
1813 * PowerPC specifics (instruction cache flushing)
1814 * ------------------------------------------------------------------------*/
1816 #ifdef powerpc_TARGET_ARCH
1818 ocFlushInstructionCache
1820 Flush the data & instruction caches.
1821 Because the PPC has split data/instruction caches, we have to
1822 do that whenever we modify code at runtime.
1825 static void ocFlushInstructionCache( ObjectCode *oc )
1827 int n = (oc->fileSize + sizeof( SymbolExtra ) * oc->n_symbol_extras + 3) / 4;
1828 unsigned long *p = (unsigned long *) oc->image;
1832 __asm__ volatile ( "dcbf 0,%0\n\t"
1840 __asm__ volatile ( "sync\n\t"
1846 /* --------------------------------------------------------------------------
1847 * PEi386 specifics (Win32 targets)
1848 * ------------------------------------------------------------------------*/
1850 /* The information for this linker comes from
1851 Microsoft Portable Executable
1852 and Common Object File Format Specification
1853 revision 5.1 January 1998
1854 which SimonM says comes from the MS Developer Network CDs.
1856 It can be found there (on older CDs), but can also be found
1859 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
1861 (this is Rev 6.0 from February 1999).
1863 Things move, so if that fails, try searching for it via
1865 http://www.google.com/search?q=PE+COFF+specification
1867 The ultimate reference for the PE format is the Winnt.h
1868 header file that comes with the Platform SDKs; as always,
1869 implementations will drift wrt their documentation.
1871 A good background article on the PE format is Matt Pietrek's
1872 March 1994 article in Microsoft System Journal (MSJ)
1873 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
1874 Win32 Portable Executable File Format." The info in there
1875 has recently been updated in a two part article in
1876 MSDN magazine, issues Feb and March 2002,
1877 "Inside Windows: An In-Depth Look into the Win32 Portable
1878 Executable File Format"
1880 John Levine's book "Linkers and Loaders" contains useful
1885 #if defined(OBJFORMAT_PEi386)
1889 typedef unsigned char UChar;
1890 typedef unsigned short UInt16;
1891 typedef unsigned int UInt32;
1898 UInt16 NumberOfSections;
1899 UInt32 TimeDateStamp;
1900 UInt32 PointerToSymbolTable;
1901 UInt32 NumberOfSymbols;
1902 UInt16 SizeOfOptionalHeader;
1903 UInt16 Characteristics;
1907 #define sizeof_COFF_header 20
1914 UInt32 VirtualAddress;
1915 UInt32 SizeOfRawData;
1916 UInt32 PointerToRawData;
1917 UInt32 PointerToRelocations;
1918 UInt32 PointerToLinenumbers;
1919 UInt16 NumberOfRelocations;
1920 UInt16 NumberOfLineNumbers;
1921 UInt32 Characteristics;
1925 #define sizeof_COFF_section 40
1932 UInt16 SectionNumber;
1935 UChar NumberOfAuxSymbols;
1939 #define sizeof_COFF_symbol 18
1944 UInt32 VirtualAddress;
1945 UInt32 SymbolTableIndex;
1950 #define sizeof_COFF_reloc 10
1953 /* From PE spec doc, section 3.3.2 */
1954 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
1955 windows.h -- for the same purpose, but I want to know what I'm
1957 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
1958 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
1959 #define MYIMAGE_FILE_DLL 0x2000
1960 #define MYIMAGE_FILE_SYSTEM 0x1000
1961 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
1962 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
1963 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
1965 /* From PE spec doc, section 5.4.2 and 5.4.4 */
1966 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
1967 #define MYIMAGE_SYM_CLASS_STATIC 3
1968 #define MYIMAGE_SYM_UNDEFINED 0
1970 /* From PE spec doc, section 4.1 */
1971 #define MYIMAGE_SCN_CNT_CODE 0x00000020
1972 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
1973 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
1975 /* From PE spec doc, section 5.2.1 */
1976 #define MYIMAGE_REL_I386_DIR32 0x0006
1977 #define MYIMAGE_REL_I386_REL32 0x0014
1980 /* We use myindex to calculate array addresses, rather than
1981 simply doing the normal subscript thing. That's because
1982 some of the above structs have sizes which are not
1983 a whole number of words. GCC rounds their sizes up to a
1984 whole number of words, which means that the address calcs
1985 arising from using normal C indexing or pointer arithmetic
1986 are just plain wrong. Sigh.
1989 myindex ( int scale, void* base, int index )
1992 ((UChar*)base) + scale * index;
1997 printName ( UChar* name, UChar* strtab )
1999 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2000 UInt32 strtab_offset = * (UInt32*)(name+4);
2001 debugBelch("%s", strtab + strtab_offset );
2004 for (i = 0; i < 8; i++) {
2005 if (name[i] == 0) break;
2006 debugBelch("%c", name[i] );
2013 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
2015 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2016 UInt32 strtab_offset = * (UInt32*)(name+4);
2017 strncpy ( dst, strtab+strtab_offset, dstSize );
2023 if (name[i] == 0) break;
2033 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
2036 /* If the string is longer than 8 bytes, look in the
2037 string table for it -- this will be correctly zero terminated.
2039 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2040 UInt32 strtab_offset = * (UInt32*)(name+4);
2041 return ((UChar*)strtab) + strtab_offset;
2043 /* Otherwise, if shorter than 8 bytes, return the original,
2044 which by defn is correctly terminated.
2046 if (name[7]==0) return name;
2047 /* The annoying case: 8 bytes. Copy into a temporary
2048 (which is never freed ...)
2050 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
2052 strncpy(newstr,name,8);
2058 /* Just compares the short names (first 8 chars) */
2059 static COFF_section *
2060 findPEi386SectionCalled ( ObjectCode* oc, char* name )
2064 = (COFF_header*)(oc->image);
2065 COFF_section* sectab
2067 ((UChar*)(oc->image))
2068 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2070 for (i = 0; i < hdr->NumberOfSections; i++) {
2073 COFF_section* section_i
2075 myindex ( sizeof_COFF_section, sectab, i );
2076 n1 = (UChar*) &(section_i->Name);
2078 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
2079 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
2080 n1[6]==n2[6] && n1[7]==n2[7])
2089 zapTrailingAtSign ( UChar* sym )
2091 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
2093 if (sym[0] == 0) return;
2095 while (sym[i] != 0) i++;
2098 while (j > 0 && my_isdigit(sym[j])) j--;
2099 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
2104 lookupSymbolInDLLs ( UChar *lbl )
2109 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
2110 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
2112 if (lbl[0] == '_') {
2113 /* HACK: if the name has an initial underscore, try stripping
2114 it off & look that up first. I've yet to verify whether there's
2115 a Rule that governs whether an initial '_' *should always* be
2116 stripped off when mapping from import lib name to the DLL name.
2118 sym = GetProcAddress(o_dll->instance, (lbl+1));
2120 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
2124 sym = GetProcAddress(o_dll->instance, lbl);
2126 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
2135 ocVerifyImage_PEi386 ( ObjectCode* oc )
2140 COFF_section* sectab;
2141 COFF_symbol* symtab;
2143 /* debugBelch("\nLOADING %s\n", oc->fileName); */
2144 hdr = (COFF_header*)(oc->image);
2145 sectab = (COFF_section*) (
2146 ((UChar*)(oc->image))
2147 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2149 symtab = (COFF_symbol*) (
2150 ((UChar*)(oc->image))
2151 + hdr->PointerToSymbolTable
2153 strtab = ((UChar*)symtab)
2154 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2156 if (hdr->Machine != 0x14c) {
2157 errorBelch("%s: Not x86 PEi386", oc->fileName);
2160 if (hdr->SizeOfOptionalHeader != 0) {
2161 errorBelch("%s: PEi386 with nonempty optional header", oc->fileName);
2164 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
2165 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
2166 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
2167 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
2168 errorBelch("%s: Not a PEi386 object file", oc->fileName);
2171 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
2172 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
2173 errorBelch("%s: Invalid PEi386 word size or endiannness: %d",
2175 (int)(hdr->Characteristics));
2178 /* If the string table size is way crazy, this might indicate that
2179 there are more than 64k relocations, despite claims to the
2180 contrary. Hence this test. */
2181 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
2183 if ( (*(UInt32*)strtab) > 600000 ) {
2184 /* Note that 600k has no special significance other than being
2185 big enough to handle the almost-2MB-sized lumps that
2186 constitute HSwin32*.o. */
2187 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
2192 /* No further verification after this point; only debug printing. */
2194 IF_DEBUG(linker, i=1);
2195 if (i == 0) return 1;
2197 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
2198 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
2199 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
2202 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
2203 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
2204 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
2205 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
2206 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
2207 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
2208 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
2210 /* Print the section table. */
2212 for (i = 0; i < hdr->NumberOfSections; i++) {
2214 COFF_section* sectab_i
2216 myindex ( sizeof_COFF_section, sectab, i );
2223 printName ( sectab_i->Name, strtab );
2233 sectab_i->VirtualSize,
2234 sectab_i->VirtualAddress,
2235 sectab_i->SizeOfRawData,
2236 sectab_i->PointerToRawData,
2237 sectab_i->NumberOfRelocations,
2238 sectab_i->PointerToRelocations,
2239 sectab_i->PointerToRawData
2241 reltab = (COFF_reloc*) (
2242 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2245 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2246 /* If the relocation field (a short) has overflowed, the
2247 * real count can be found in the first reloc entry.
2249 * See Section 4.1 (last para) of the PE spec (rev6.0).
2251 COFF_reloc* rel = (COFF_reloc*)
2252 myindex ( sizeof_COFF_reloc, reltab, 0 );
2253 noRelocs = rel->VirtualAddress;
2256 noRelocs = sectab_i->NumberOfRelocations;
2260 for (; j < noRelocs; j++) {
2262 COFF_reloc* rel = (COFF_reloc*)
2263 myindex ( sizeof_COFF_reloc, reltab, j );
2265 " type 0x%-4x vaddr 0x%-8x name `",
2267 rel->VirtualAddress );
2268 sym = (COFF_symbol*)
2269 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
2270 /* Hmm..mysterious looking offset - what's it for? SOF */
2271 printName ( sym->Name, strtab -10 );
2278 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
2279 debugBelch("---START of string table---\n");
2280 for (i = 4; i < *(Int32*)strtab; i++) {
2282 debugBelch("\n"); else
2283 debugBelch("%c", strtab[i] );
2285 debugBelch("--- END of string table---\n");
2290 COFF_symbol* symtab_i;
2291 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2292 symtab_i = (COFF_symbol*)
2293 myindex ( sizeof_COFF_symbol, symtab, i );
2299 printName ( symtab_i->Name, strtab );
2308 (Int32)(symtab_i->SectionNumber),
2309 (UInt32)symtab_i->Type,
2310 (UInt32)symtab_i->StorageClass,
2311 (UInt32)symtab_i->NumberOfAuxSymbols
2313 i += symtab_i->NumberOfAuxSymbols;
2323 ocGetNames_PEi386 ( ObjectCode* oc )
2326 COFF_section* sectab;
2327 COFF_symbol* symtab;
2334 hdr = (COFF_header*)(oc->image);
2335 sectab = (COFF_section*) (
2336 ((UChar*)(oc->image))
2337 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2339 symtab = (COFF_symbol*) (
2340 ((UChar*)(oc->image))
2341 + hdr->PointerToSymbolTable
2343 strtab = ((UChar*)(oc->image))
2344 + hdr->PointerToSymbolTable
2345 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2347 /* Allocate space for any (local, anonymous) .bss sections. */
2349 for (i = 0; i < hdr->NumberOfSections; i++) {
2352 COFF_section* sectab_i
2354 myindex ( sizeof_COFF_section, sectab, i );
2355 if (0 != strcmp(sectab_i->Name, ".bss")) continue;
2356 /* sof 10/05: the PE spec text isn't too clear regarding what
2357 * the SizeOfRawData field is supposed to hold for object
2358 * file sections containing just uninitialized data -- for executables,
2359 * it is supposed to be zero; unclear what it's supposed to be
2360 * for object files. However, VirtualSize is guaranteed to be
2361 * zero for object files, which definitely suggests that SizeOfRawData
2362 * will be non-zero (where else would the size of this .bss section be
2363 * stored?) Looking at the COFF_section info for incoming object files,
2364 * this certainly appears to be the case.
2366 * => I suspect we've been incorrectly handling .bss sections in (relocatable)
2367 * object files up until now. This turned out to bite us with ghc-6.4.1's use
2368 * of gcc-3.4.x, which has started to emit initially-zeroed-out local 'static'
2369 * variable decls into to the .bss section. (The specific function in Q which
2370 * triggered this is libraries/base/cbits/dirUtils.c:__hscore_getFolderPath())
2372 if (sectab_i->VirtualSize == 0 && sectab_i->SizeOfRawData == 0) continue;
2373 /* This is a non-empty .bss section. Allocate zeroed space for
2374 it, and set its PointerToRawData field such that oc->image +
2375 PointerToRawData == addr_of_zeroed_space. */
2376 bss_sz = sectab_i->VirtualSize;
2377 if ( bss_sz < sectab_i->SizeOfRawData) { bss_sz = sectab_i->SizeOfRawData; }
2378 zspace = stgCallocBytes(1, bss_sz, "ocGetNames_PEi386(anonymous bss)");
2379 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
2380 addProddableBlock(oc, zspace, bss_sz);
2381 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
2384 /* Copy section information into the ObjectCode. */
2386 for (i = 0; i < hdr->NumberOfSections; i++) {
2392 = SECTIONKIND_OTHER;
2393 COFF_section* sectab_i
2395 myindex ( sizeof_COFF_section, sectab, i );
2396 IF_DEBUG(linker, debugBelch("section name = %s\n", sectab_i->Name ));
2399 /* I'm sure this is the Right Way to do it. However, the
2400 alternative of testing the sectab_i->Name field seems to
2401 work ok with Cygwin.
2403 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
2404 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
2405 kind = SECTIONKIND_CODE_OR_RODATA;
2408 if (0==strcmp(".text",sectab_i->Name) ||
2409 0==strcmp(".rdata",sectab_i->Name)||
2410 0==strcmp(".rodata",sectab_i->Name))
2411 kind = SECTIONKIND_CODE_OR_RODATA;
2412 if (0==strcmp(".data",sectab_i->Name) ||
2413 0==strcmp(".bss",sectab_i->Name))
2414 kind = SECTIONKIND_RWDATA;
2416 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
2417 sz = sectab_i->SizeOfRawData;
2418 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
2420 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
2421 end = start + sz - 1;
2423 if (kind == SECTIONKIND_OTHER
2424 /* Ignore sections called which contain stabs debugging
2426 && 0 != strcmp(".stab", sectab_i->Name)
2427 && 0 != strcmp(".stabstr", sectab_i->Name)
2428 /* ignore constructor section for now */
2429 && 0 != strcmp(".ctors", sectab_i->Name)
2430 /* ignore section generated from .ident */
2431 && 0!= strcmp("/4", sectab_i->Name)
2432 /* ignore unknown section that appeared in gcc 3.4.5(?) */
2433 && 0!= strcmp(".reloc", sectab_i->Name)
2435 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName);
2439 if (kind != SECTIONKIND_OTHER && end >= start) {
2440 addSection(oc, kind, start, end);
2441 addProddableBlock(oc, start, end - start + 1);
2445 /* Copy exported symbols into the ObjectCode. */
2447 oc->n_symbols = hdr->NumberOfSymbols;
2448 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2449 "ocGetNames_PEi386(oc->symbols)");
2450 /* Call me paranoid; I don't care. */
2451 for (i = 0; i < oc->n_symbols; i++)
2452 oc->symbols[i] = NULL;
2456 COFF_symbol* symtab_i;
2457 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2458 symtab_i = (COFF_symbol*)
2459 myindex ( sizeof_COFF_symbol, symtab, i );
2463 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2464 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2465 /* This symbol is global and defined, viz, exported */
2466 /* for MYIMAGE_SYMCLASS_EXTERNAL
2467 && !MYIMAGE_SYM_UNDEFINED,
2468 the address of the symbol is:
2469 address of relevant section + offset in section
2471 COFF_section* sectabent
2472 = (COFF_section*) myindex ( sizeof_COFF_section,
2474 symtab_i->SectionNumber-1 );
2475 addr = ((UChar*)(oc->image))
2476 + (sectabent->PointerToRawData
2480 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2481 && symtab_i->Value > 0) {
2482 /* This symbol isn't in any section at all, ie, global bss.
2483 Allocate zeroed space for it. */
2484 addr = stgCallocBytes(1, symtab_i->Value,
2485 "ocGetNames_PEi386(non-anonymous bss)");
2486 addSection(oc, SECTIONKIND_RWDATA, addr,
2487 ((UChar*)addr) + symtab_i->Value - 1);
2488 addProddableBlock(oc, addr, symtab_i->Value);
2489 /* debugBelch("BSS section at 0x%x\n", addr); */
2492 if (addr != NULL ) {
2493 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2494 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2495 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2496 ASSERT(i >= 0 && i < oc->n_symbols);
2497 /* cstring_from_COFF_symbol_name always succeeds. */
2498 oc->symbols[i] = sname;
2499 ghciInsertStrHashTable(oc->fileName, symhash, sname, addr);
2503 "IGNORING symbol %d\n"
2507 printName ( symtab_i->Name, strtab );
2516 (Int32)(symtab_i->SectionNumber),
2517 (UInt32)symtab_i->Type,
2518 (UInt32)symtab_i->StorageClass,
2519 (UInt32)symtab_i->NumberOfAuxSymbols
2524 i += symtab_i->NumberOfAuxSymbols;
2533 ocResolve_PEi386 ( ObjectCode* oc )
2536 COFF_section* sectab;
2537 COFF_symbol* symtab;
2547 /* ToDo: should be variable-sized? But is at least safe in the
2548 sense of buffer-overrun-proof. */
2550 /* debugBelch("resolving for %s\n", oc->fileName); */
2552 hdr = (COFF_header*)(oc->image);
2553 sectab = (COFF_section*) (
2554 ((UChar*)(oc->image))
2555 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2557 symtab = (COFF_symbol*) (
2558 ((UChar*)(oc->image))
2559 + hdr->PointerToSymbolTable
2561 strtab = ((UChar*)(oc->image))
2562 + hdr->PointerToSymbolTable
2563 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2565 for (i = 0; i < hdr->NumberOfSections; i++) {
2566 COFF_section* sectab_i
2568 myindex ( sizeof_COFF_section, sectab, i );
2571 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2574 /* Ignore sections called which contain stabs debugging
2576 if (0 == strcmp(".stab", sectab_i->Name)
2577 || 0 == strcmp(".stabstr", sectab_i->Name)
2578 || 0 == strcmp(".ctors", sectab_i->Name))
2581 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2582 /* If the relocation field (a short) has overflowed, the
2583 * real count can be found in the first reloc entry.
2585 * See Section 4.1 (last para) of the PE spec (rev6.0).
2587 * Nov2003 update: the GNU linker still doesn't correctly
2588 * handle the generation of relocatable object files with
2589 * overflown relocations. Hence the output to warn of potential
2592 COFF_reloc* rel = (COFF_reloc*)
2593 myindex ( sizeof_COFF_reloc, reltab, 0 );
2594 noRelocs = rel->VirtualAddress;
2596 /* 10/05: we now assume (and check for) a GNU ld that is capable
2597 * of handling object files with (>2^16) of relocs.
2600 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2605 noRelocs = sectab_i->NumberOfRelocations;
2610 for (; j < noRelocs; j++) {
2612 COFF_reloc* reltab_j
2614 myindex ( sizeof_COFF_reloc, reltab, j );
2616 /* the location to patch */
2618 ((UChar*)(oc->image))
2619 + (sectab_i->PointerToRawData
2620 + reltab_j->VirtualAddress
2621 - sectab_i->VirtualAddress )
2623 /* the existing contents of pP */
2625 /* the symbol to connect to */
2626 sym = (COFF_symbol*)
2627 myindex ( sizeof_COFF_symbol,
2628 symtab, reltab_j->SymbolTableIndex );
2631 "reloc sec %2d num %3d: type 0x%-4x "
2632 "vaddr 0x%-8x name `",
2634 (UInt32)reltab_j->Type,
2635 reltab_j->VirtualAddress );
2636 printName ( sym->Name, strtab );
2637 debugBelch("'\n" ));
2639 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2640 COFF_section* section_sym
2641 = findPEi386SectionCalled ( oc, sym->Name );
2643 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2646 S = ((UInt32)(oc->image))
2647 + (section_sym->PointerToRawData
2650 copyName ( sym->Name, strtab, symbol, 1000-1 );
2651 S = (UInt32) lookupSymbol( symbol );
2652 if ((void*)S != NULL) goto foundit;
2653 /* Newline first because the interactive linker has printed "linking..." */
2654 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, symbol);
2658 checkProddableBlock(oc, pP);
2659 switch (reltab_j->Type) {
2660 case MYIMAGE_REL_I386_DIR32:
2663 case MYIMAGE_REL_I386_REL32:
2664 /* Tricky. We have to insert a displacement at
2665 pP which, when added to the PC for the _next_
2666 insn, gives the address of the target (S).
2667 Problem is to know the address of the next insn
2668 when we only know pP. We assume that this
2669 literal field is always the last in the insn,
2670 so that the address of the next insn is pP+4
2671 -- hence the constant 4.
2672 Also I don't know if A should be added, but so
2673 far it has always been zero.
2675 SOF 05/2005: 'A' (old contents of *pP) have been observed
2676 to contain values other than zero (the 'wx' object file
2677 that came with wxhaskell-0.9.4; dunno how it was compiled..).
2678 So, add displacement to old value instead of asserting
2679 A to be zero. Fixes wxhaskell-related crashes, and no other
2680 ill effects have been observed.
2682 Update: the reason why we're seeing these more elaborate
2683 relocations is due to a switch in how the NCG compiles SRTs
2684 and offsets to them from info tables. SRTs live in .(ro)data,
2685 while info tables live in .text, causing GAS to emit REL32/DISP32
2686 relocations with non-zero values. Adding the displacement is
2687 the right thing to do.
2689 *pP = S - ((UInt32)pP) - 4 + A;
2692 debugBelch("%s: unhandled PEi386 relocation type %d",
2693 oc->fileName, reltab_j->Type);
2700 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2704 #endif /* defined(OBJFORMAT_PEi386) */
2707 /* --------------------------------------------------------------------------
2709 * ------------------------------------------------------------------------*/
2711 #if defined(OBJFORMAT_ELF)
2716 #if defined(sparc_HOST_ARCH)
2717 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2718 #elif defined(i386_HOST_ARCH)
2719 # define ELF_TARGET_386 /* Used inside <elf.h> */
2720 #elif defined(x86_64_HOST_ARCH)
2721 # define ELF_TARGET_X64_64
2723 #elif defined (ia64_HOST_ARCH)
2724 # define ELF_TARGET_IA64 /* Used inside <elf.h> */
2726 # define ELF_FUNCTION_DESC /* calling convention uses function descriptors */
2727 # define ELF_NEED_GOT /* needs Global Offset Table */
2728 # define ELF_NEED_PLT /* needs Procedure Linkage Tables */
2731 #if !defined(openbsd_HOST_OS)
2734 /* openbsd elf has things in different places, with diff names */
2735 # include <elf_abi.h>
2736 # include <machine/reloc.h>
2737 # define R_386_32 RELOC_32
2738 # define R_386_PC32 RELOC_PC32
2741 /* If elf.h doesn't define it */
2742 # ifndef R_X86_64_PC64
2743 # define R_X86_64_PC64 24
2747 * Define a set of types which can be used for both ELF32 and ELF64
2751 #define ELFCLASS ELFCLASS64
2752 #define Elf_Addr Elf64_Addr
2753 #define Elf_Word Elf64_Word
2754 #define Elf_Sword Elf64_Sword
2755 #define Elf_Ehdr Elf64_Ehdr
2756 #define Elf_Phdr Elf64_Phdr
2757 #define Elf_Shdr Elf64_Shdr
2758 #define Elf_Sym Elf64_Sym
2759 #define Elf_Rel Elf64_Rel
2760 #define Elf_Rela Elf64_Rela
2761 #define ELF_ST_TYPE ELF64_ST_TYPE
2762 #define ELF_ST_BIND ELF64_ST_BIND
2763 #define ELF_R_TYPE ELF64_R_TYPE
2764 #define ELF_R_SYM ELF64_R_SYM
2766 #define ELFCLASS ELFCLASS32
2767 #define Elf_Addr Elf32_Addr
2768 #define Elf_Word Elf32_Word
2769 #define Elf_Sword Elf32_Sword
2770 #define Elf_Ehdr Elf32_Ehdr
2771 #define Elf_Phdr Elf32_Phdr
2772 #define Elf_Shdr Elf32_Shdr
2773 #define Elf_Sym Elf32_Sym
2774 #define Elf_Rel Elf32_Rel
2775 #define Elf_Rela Elf32_Rela
2777 #define ELF_ST_TYPE ELF32_ST_TYPE
2780 #define ELF_ST_BIND ELF32_ST_BIND
2783 #define ELF_R_TYPE ELF32_R_TYPE
2786 #define ELF_R_SYM ELF32_R_SYM
2792 * Functions to allocate entries in dynamic sections. Currently we simply
2793 * preallocate a large number, and we don't check if a entry for the given
2794 * target already exists (a linear search is too slow). Ideally these
2795 * entries would be associated with symbols.
2798 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
2799 #define GOT_SIZE 0x20000
2800 #define FUNCTION_TABLE_SIZE 0x10000
2801 #define PLT_SIZE 0x08000
2804 static Elf_Addr got[GOT_SIZE];
2805 static unsigned int gotIndex;
2806 static Elf_Addr gp_val = (Elf_Addr)got;
2809 allocateGOTEntry(Elf_Addr target)
2813 if (gotIndex >= GOT_SIZE)
2814 barf("Global offset table overflow");
2816 entry = &got[gotIndex++];
2818 return (Elf_Addr)entry;
2822 #ifdef ELF_FUNCTION_DESC
2828 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
2829 static unsigned int functionTableIndex;
2832 allocateFunctionDesc(Elf_Addr target)
2834 FunctionDesc *entry;
2836 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
2837 barf("Function table overflow");
2839 entry = &functionTable[functionTableIndex++];
2841 entry->gp = (Elf_Addr)gp_val;
2842 return (Elf_Addr)entry;
2846 copyFunctionDesc(Elf_Addr target)
2848 FunctionDesc *olddesc = (FunctionDesc *)target;
2849 FunctionDesc *newdesc;
2851 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
2852 newdesc->gp = olddesc->gp;
2853 return (Elf_Addr)newdesc;
2858 #ifdef ia64_HOST_ARCH
2859 static void ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value);
2860 static void ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc);
2862 static unsigned char plt_code[] =
2864 /* taken from binutils bfd/elfxx-ia64.c */
2865 0x0b, 0x78, 0x00, 0x02, 0x00, 0x24, /* [MMI] addl r15=0,r1;; */
2866 0x00, 0x41, 0x3c, 0x30, 0x28, 0xc0, /* ld8 r16=[r15],8 */
2867 0x01, 0x08, 0x00, 0x84, /* mov r14=r1;; */
2868 0x11, 0x08, 0x00, 0x1e, 0x18, 0x10, /* [MIB] ld8 r1=[r15] */
2869 0x60, 0x80, 0x04, 0x80, 0x03, 0x00, /* mov b6=r16 */
2870 0x60, 0x00, 0x80, 0x00 /* br.few b6;; */
2873 /* If we can't get to the function descriptor via gp, take a local copy of it */
2874 #define PLT_RELOC(code, target) { \
2875 Elf64_Sxword rel_value = target - gp_val; \
2876 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff)) \
2877 ia64_reloc_gprel22((Elf_Addr)code, copyFunctionDesc(target)); \
2879 ia64_reloc_gprel22((Elf_Addr)code, target); \
2884 unsigned char code[sizeof(plt_code)];
2888 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
2890 PLTEntry *plt = (PLTEntry *)oc->plt;
2893 if (oc->pltIndex >= PLT_SIZE)
2894 barf("Procedure table overflow");
2896 entry = &plt[oc->pltIndex++];
2897 memcpy(entry->code, plt_code, sizeof(entry->code));
2898 PLT_RELOC(entry->code, target);
2899 return (Elf_Addr)entry;
2905 return (PLT_SIZE * sizeof(PLTEntry));
2911 * Generic ELF functions
2915 findElfSection ( void* objImage, Elf_Word sh_type )
2917 char* ehdrC = (char*)objImage;
2918 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2919 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
2920 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2924 for (i = 0; i < ehdr->e_shnum; i++) {
2925 if (shdr[i].sh_type == sh_type
2926 /* Ignore the section header's string table. */
2927 && i != ehdr->e_shstrndx
2928 /* Ignore string tables named .stabstr, as they contain
2930 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2932 ptr = ehdrC + shdr[i].sh_offset;
2939 #if defined(ia64_HOST_ARCH)
2941 findElfSegment ( void* objImage, Elf_Addr vaddr )
2943 char* ehdrC = (char*)objImage;
2944 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2945 Elf_Phdr* phdr = (Elf_Phdr*)(ehdrC + ehdr->e_phoff);
2946 Elf_Addr segaddr = 0;
2949 for (i = 0; i < ehdr->e_phnum; i++) {
2950 segaddr = phdr[i].p_vaddr;
2951 if ((vaddr >= segaddr) && (vaddr < segaddr + phdr[i].p_memsz))
2959 ocVerifyImage_ELF ( ObjectCode* oc )
2963 int i, j, nent, nstrtab, nsymtabs;
2967 char* ehdrC = (char*)(oc->image);
2968 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2970 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
2971 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
2972 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
2973 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
2974 errorBelch("%s: not an ELF object", oc->fileName);
2978 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
2979 errorBelch("%s: unsupported ELF format", oc->fileName);
2983 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
2984 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
2986 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
2987 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
2989 errorBelch("%s: unknown endiannness", oc->fileName);
2993 if (ehdr->e_type != ET_REL) {
2994 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
2997 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
2999 IF_DEBUG(linker,debugBelch( "Architecture is " ));
3000 switch (ehdr->e_machine) {
3001 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
3002 #ifdef EM_SPARC32PLUS
3003 case EM_SPARC32PLUS:
3005 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
3007 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
3009 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
3011 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
3012 #elif defined(EM_AMD64)
3013 case EM_AMD64: IF_DEBUG(linker,debugBelch( "amd64" )); break;
3015 default: IF_DEBUG(linker,debugBelch( "unknown" ));
3016 errorBelch("%s: unknown architecture (e_machine == %d)"
3017 , oc->fileName, ehdr->e_machine);
3021 IF_DEBUG(linker,debugBelch(
3022 "\nSection header table: start %ld, n_entries %d, ent_size %d\n",
3023 (long)ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
3025 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
3027 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3029 if (ehdr->e_shstrndx == SHN_UNDEF) {
3030 errorBelch("%s: no section header string table", oc->fileName);
3033 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
3035 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3038 for (i = 0; i < ehdr->e_shnum; i++) {
3039 IF_DEBUG(linker,debugBelch("%2d: ", i ));
3040 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
3041 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
3042 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
3043 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
3044 ehdrC + shdr[i].sh_offset,
3045 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
3047 if (shdr[i].sh_type == SHT_REL) {
3048 IF_DEBUG(linker,debugBelch("Rel " ));
3049 } else if (shdr[i].sh_type == SHT_RELA) {
3050 IF_DEBUG(linker,debugBelch("RelA " ));
3052 IF_DEBUG(linker,debugBelch(" "));
3055 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
3059 IF_DEBUG(linker,debugBelch( "\nString tables" ));
3062 for (i = 0; i < ehdr->e_shnum; i++) {
3063 if (shdr[i].sh_type == SHT_STRTAB
3064 /* Ignore the section header's string table. */
3065 && i != ehdr->e_shstrndx
3066 /* Ignore string tables named .stabstr, as they contain
3068 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3070 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
3071 strtab = ehdrC + shdr[i].sh_offset;
3076 errorBelch("%s: no string tables, or too many", oc->fileName);
3081 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
3082 for (i = 0; i < ehdr->e_shnum; i++) {
3083 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3084 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
3086 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3087 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3088 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%ld rem)\n",
3090 (long)shdr[i].sh_size % sizeof(Elf_Sym)
3092 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
3093 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
3096 for (j = 0; j < nent; j++) {
3097 IF_DEBUG(linker,debugBelch(" %2d ", j ));
3098 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
3099 (int)stab[j].st_shndx,
3100 (int)stab[j].st_size,
3101 (char*)stab[j].st_value ));
3103 IF_DEBUG(linker,debugBelch("type=" ));
3104 switch (ELF_ST_TYPE(stab[j].st_info)) {
3105 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
3106 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
3107 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
3108 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
3109 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
3110 default: IF_DEBUG(linker,debugBelch("? " )); break;
3112 IF_DEBUG(linker,debugBelch(" " ));
3114 IF_DEBUG(linker,debugBelch("bind=" ));
3115 switch (ELF_ST_BIND(stab[j].st_info)) {
3116 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
3117 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
3118 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
3119 default: IF_DEBUG(linker,debugBelch("? " )); break;
3121 IF_DEBUG(linker,debugBelch(" " ));
3123 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
3127 if (nsymtabs == 0) {
3128 errorBelch("%s: didn't find any symbol tables", oc->fileName);
3135 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
3139 if (hdr->sh_type == SHT_PROGBITS
3140 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
3141 /* .text-style section */
3142 return SECTIONKIND_CODE_OR_RODATA;
3145 if (hdr->sh_type == SHT_PROGBITS
3146 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3147 /* .data-style section */
3148 return SECTIONKIND_RWDATA;
3151 if (hdr->sh_type == SHT_PROGBITS
3152 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
3153 /* .rodata-style section */
3154 return SECTIONKIND_CODE_OR_RODATA;
3157 if (hdr->sh_type == SHT_NOBITS
3158 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3159 /* .bss-style section */
3161 return SECTIONKIND_RWDATA;
3164 return SECTIONKIND_OTHER;
3169 ocGetNames_ELF ( ObjectCode* oc )
3174 char* ehdrC = (char*)(oc->image);
3175 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3176 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
3177 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3179 ASSERT(symhash != NULL);
3182 errorBelch("%s: no strtab", oc->fileName);
3187 for (i = 0; i < ehdr->e_shnum; i++) {
3188 /* Figure out what kind of section it is. Logic derived from
3189 Figure 1.14 ("Special Sections") of the ELF document
3190 ("Portable Formats Specification, Version 1.1"). */
3192 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
3194 if (is_bss && shdr[i].sh_size > 0) {
3195 /* This is a non-empty .bss section. Allocate zeroed space for
3196 it, and set its .sh_offset field such that
3197 ehdrC + .sh_offset == addr_of_zeroed_space. */
3198 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
3199 "ocGetNames_ELF(BSS)");
3200 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
3202 debugBelch("BSS section at 0x%x, size %d\n",
3203 zspace, shdr[i].sh_size);
3207 /* fill in the section info */
3208 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
3209 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
3210 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
3211 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
3214 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3216 /* copy stuff into this module's object symbol table */
3217 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3218 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3220 oc->n_symbols = nent;
3221 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3222 "ocGetNames_ELF(oc->symbols)");
3224 for (j = 0; j < nent; j++) {
3226 char isLocal = FALSE; /* avoids uninit-var warning */
3228 char* nm = strtab + stab[j].st_name;
3229 int secno = stab[j].st_shndx;
3231 /* Figure out if we want to add it; if so, set ad to its
3232 address. Otherwise leave ad == NULL. */
3234 if (secno == SHN_COMMON) {
3236 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
3238 debugBelch("COMMON symbol, size %d name %s\n",
3239 stab[j].st_size, nm);
3241 /* Pointless to do addProddableBlock() for this area,
3242 since the linker should never poke around in it. */
3245 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
3246 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
3248 /* and not an undefined symbol */
3249 && stab[j].st_shndx != SHN_UNDEF
3250 /* and not in a "special section" */
3251 && stab[j].st_shndx < SHN_LORESERVE
3253 /* and it's a not a section or string table or anything silly */
3254 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
3255 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
3256 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
3259 /* Section 0 is the undefined section, hence > and not >=. */
3260 ASSERT(secno > 0 && secno < ehdr->e_shnum);
3262 if (shdr[secno].sh_type == SHT_NOBITS) {
3263 debugBelch(" BSS symbol, size %d off %d name %s\n",
3264 stab[j].st_size, stab[j].st_value, nm);
3267 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
3268 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
3271 #ifdef ELF_FUNCTION_DESC
3272 /* dlsym() and the initialisation table both give us function
3273 * descriptors, so to be consistent we store function descriptors
3274 * in the symbol table */
3275 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
3276 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
3278 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s\n",
3279 ad, oc->fileName, nm ));
3284 /* And the decision is ... */
3288 oc->symbols[j] = nm;
3291 /* Ignore entirely. */
3293 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
3297 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
3298 strtab + stab[j].st_name ));
3301 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
3302 (int)ELF_ST_BIND(stab[j].st_info),
3303 (int)ELF_ST_TYPE(stab[j].st_info),
3304 (int)stab[j].st_shndx,
3305 strtab + stab[j].st_name
3308 oc->symbols[j] = NULL;
3317 /* Do ELF relocations which lack an explicit addend. All x86-linux
3318 relocations appear to be of this form. */
3320 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
3321 Elf_Shdr* shdr, int shnum,
3322 Elf_Sym* stab, char* strtab )
3327 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
3328 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
3329 int target_shndx = shdr[shnum].sh_info;
3330 int symtab_shndx = shdr[shnum].sh_link;
3332 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3333 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
3334 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3335 target_shndx, symtab_shndx ));
3337 /* Skip sections that we're not interested in. */
3340 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
3341 if (kind == SECTIONKIND_OTHER) {
3342 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
3347 for (j = 0; j < nent; j++) {
3348 Elf_Addr offset = rtab[j].r_offset;
3349 Elf_Addr info = rtab[j].r_info;
3351 Elf_Addr P = ((Elf_Addr)targ) + offset;
3352 Elf_Word* pP = (Elf_Word*)P;
3357 StgStablePtr stablePtr;
3360 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
3361 j, (void*)offset, (void*)info ));
3363 IF_DEBUG(linker,debugBelch( " ZERO" ));
3366 Elf_Sym sym = stab[ELF_R_SYM(info)];
3367 /* First see if it is a local symbol. */
3368 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3369 /* Yes, so we can get the address directly from the ELF symbol
3371 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3373 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3374 + stab[ELF_R_SYM(info)].st_value);
3377 symbol = strtab + sym.st_name;
3378 stablePtr = (StgStablePtr)lookupHashTable(stablehash, (StgWord)symbol);
3379 if (NULL == stablePtr) {
3380 /* No, so look up the name in our global table. */
3381 S_tmp = lookupSymbol( symbol );
3382 S = (Elf_Addr)S_tmp;
3384 stableVal = deRefStablePtr( stablePtr );
3386 S = (Elf_Addr)S_tmp;
3390 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3393 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
3396 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
3397 (void*)P, (void*)S, (void*)A ));
3398 checkProddableBlock ( oc, pP );
3402 switch (ELF_R_TYPE(info)) {
3403 # ifdef i386_HOST_ARCH
3404 case R_386_32: *pP = value; break;
3405 case R_386_PC32: *pP = value - P; break;
3408 errorBelch("%s: unhandled ELF relocation(Rel) type %lu\n",
3409 oc->fileName, (lnat)ELF_R_TYPE(info));
3417 /* Do ELF relocations for which explicit addends are supplied.
3418 sparc-solaris relocations appear to be of this form. */
3420 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
3421 Elf_Shdr* shdr, int shnum,
3422 Elf_Sym* stab, char* strtab )
3425 char *symbol = NULL;
3427 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
3428 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
3429 int target_shndx = shdr[shnum].sh_info;
3430 int symtab_shndx = shdr[shnum].sh_link;
3432 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3433 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
3434 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3435 target_shndx, symtab_shndx ));
3437 for (j = 0; j < nent; j++) {
3438 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3439 /* This #ifdef only serves to avoid unused-var warnings. */
3440 Elf_Addr offset = rtab[j].r_offset;
3441 Elf_Addr P = targ + offset;
3443 Elf_Addr info = rtab[j].r_info;
3444 Elf_Addr A = rtab[j].r_addend;
3448 # if defined(sparc_HOST_ARCH)
3449 Elf_Word* pP = (Elf_Word*)P;
3451 # elif defined(ia64_HOST_ARCH)
3452 Elf64_Xword *pP = (Elf64_Xword *)P;
3454 # elif defined(powerpc_HOST_ARCH)
3458 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
3459 j, (void*)offset, (void*)info,
3462 IF_DEBUG(linker,debugBelch( " ZERO" ));
3465 Elf_Sym sym = stab[ELF_R_SYM(info)];
3466 /* First see if it is a local symbol. */
3467 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3468 /* Yes, so we can get the address directly from the ELF symbol
3470 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3472 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3473 + stab[ELF_R_SYM(info)].st_value);
3474 #ifdef ELF_FUNCTION_DESC
3475 /* Make a function descriptor for this function */
3476 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
3477 S = allocateFunctionDesc(S + A);
3482 /* No, so look up the name in our global table. */
3483 symbol = strtab + sym.st_name;
3484 S_tmp = lookupSymbol( symbol );
3485 S = (Elf_Addr)S_tmp;
3487 #ifdef ELF_FUNCTION_DESC
3488 /* If a function, already a function descriptor - we would
3489 have to copy it to add an offset. */
3490 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3491 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3495 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3498 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3501 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3502 (void*)P, (void*)S, (void*)A ));
3503 /* checkProddableBlock ( oc, (void*)P ); */
3507 switch (ELF_R_TYPE(info)) {
3508 # if defined(sparc_HOST_ARCH)
3509 case R_SPARC_WDISP30:
3510 w1 = *pP & 0xC0000000;
3511 w2 = (Elf_Word)((value - P) >> 2);
3512 ASSERT((w2 & 0xC0000000) == 0);
3517 w1 = *pP & 0xFFC00000;
3518 w2 = (Elf_Word)(value >> 10);
3519 ASSERT((w2 & 0xFFC00000) == 0);
3525 w2 = (Elf_Word)(value & 0x3FF);
3526 ASSERT((w2 & ~0x3FF) == 0);
3530 /* According to the Sun documentation:
3532 This relocation type resembles R_SPARC_32, except it refers to an
3533 unaligned word. That is, the word to be relocated must be treated
3534 as four separate bytes with arbitrary alignment, not as a word
3535 aligned according to the architecture requirements.
3537 (JRS: which means that freeloading on the R_SPARC_32 case
3538 is probably wrong, but hey ...)
3542 w2 = (Elf_Word)value;
3545 # elif defined(ia64_HOST_ARCH)
3546 case R_IA64_DIR64LSB:
3547 case R_IA64_FPTR64LSB:
3550 case R_IA64_PCREL64LSB:
3553 case R_IA64_SEGREL64LSB:
3554 addr = findElfSegment(ehdrC, value);
3557 case R_IA64_GPREL22:
3558 ia64_reloc_gprel22(P, value);
3560 case R_IA64_LTOFF22:
3561 case R_IA64_LTOFF22X:
3562 case R_IA64_LTOFF_FPTR22:
3563 addr = allocateGOTEntry(value);
3564 ia64_reloc_gprel22(P, addr);
3566 case R_IA64_PCREL21B:
3567 ia64_reloc_pcrel21(P, S, oc);
3570 /* This goes with R_IA64_LTOFF22X and points to the load to
3571 * convert into a move. We don't implement relaxation. */
3573 # elif defined(powerpc_HOST_ARCH)
3574 case R_PPC_ADDR16_LO:
3575 *(Elf32_Half*) P = value;
3578 case R_PPC_ADDR16_HI:
3579 *(Elf32_Half*) P = value >> 16;
3582 case R_PPC_ADDR16_HA:
3583 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3587 *(Elf32_Word *) P = value;
3591 *(Elf32_Word *) P = value - P;
3597 if( delta << 6 >> 6 != delta )
3599 value = (Elf_Addr) (&makeSymbolExtra( oc, ELF_R_SYM(info), value )
3603 if( value == 0 || delta << 6 >> 6 != delta )
3605 barf( "Unable to make SymbolExtra for #%d",
3611 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3612 | (delta & 0x3fffffc);
3616 #if x86_64_HOST_ARCH
3618 *(Elf64_Xword *)P = value;
3623 StgInt64 off = value - P;
3624 if (off >= 0x7fffffffL || off < -0x80000000L) {
3625 #if X86_64_ELF_NONPIC_HACK
3626 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3628 off = pltAddress + A - P;
3630 barf("R_X86_64_PC32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3631 symbol, off, oc->fileName );
3634 *(Elf64_Word *)P = (Elf64_Word)off;
3640 StgInt64 off = value - P;
3641 *(Elf64_Word *)P = (Elf64_Word)off;
3646 if (value >= 0x7fffffffL) {
3647 #if X86_64_ELF_NONPIC_HACK
3648 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3650 value = pltAddress + A;
3652 barf("R_X86_64_32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3653 symbol, value, oc->fileName );
3656 *(Elf64_Word *)P = (Elf64_Word)value;
3660 if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) {
3661 #if X86_64_ELF_NONPIC_HACK
3662 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3664 value = pltAddress + A;
3666 barf("R_X86_64_32S relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3667 symbol, value, oc->fileName );
3670 *(Elf64_Sword *)P = (Elf64_Sword)value;
3673 case R_X86_64_GOTPCREL:
3675 StgInt64 gotAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)->addr;
3676 StgInt64 off = gotAddress + A - P;
3677 *(Elf64_Word *)P = (Elf64_Word)off;
3681 case R_X86_64_PLT32:
3683 StgInt64 off = value - P;
3684 if (off >= 0x7fffffffL || off < -0x80000000L) {
3685 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3687 off = pltAddress + A - P;
3689 *(Elf64_Word *)P = (Elf64_Word)off;
3695 errorBelch("%s: unhandled ELF relocation(RelA) type %lu\n",
3696 oc->fileName, (lnat)ELF_R_TYPE(info));
3705 ocResolve_ELF ( ObjectCode* oc )
3709 Elf_Sym* stab = NULL;
3710 char* ehdrC = (char*)(oc->image);
3711 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3712 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3714 /* first find "the" symbol table */
3715 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3717 /* also go find the string table */
3718 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3720 if (stab == NULL || strtab == NULL) {
3721 errorBelch("%s: can't find string or symbol table", oc->fileName);
3725 /* Process the relocation sections. */
3726 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3727 if (shdr[shnum].sh_type == SHT_REL) {
3728 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3729 shnum, stab, strtab );
3733 if (shdr[shnum].sh_type == SHT_RELA) {
3734 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3735 shnum, stab, strtab );
3740 #if defined(powerpc_HOST_ARCH)
3741 ocFlushInstructionCache( oc );
3749 * Instructions are 41 bits long, packed into 128 bit bundles with a 5-bit template
3750 * at the front. The following utility functions pack and unpack instructions, and
3751 * take care of the most common relocations.
3754 #ifdef ia64_HOST_ARCH
3757 ia64_extract_instruction(Elf64_Xword *target)
3760 int slot = (Elf_Addr)target & 3;
3761 target = (Elf_Addr)target & ~3;
3769 return ((w1 >> 5) & 0x1ffffffffff);
3771 return (w1 >> 46) | ((w2 & 0x7fffff) << 18);
3775 barf("ia64_extract_instruction: invalid slot %p", target);
3780 ia64_deposit_instruction(Elf64_Xword *target, Elf64_Xword value)
3782 int slot = (Elf_Addr)target & 3;
3783 target = (Elf_Addr)target & ~3;
3788 *target |= value << 5;
3791 *target |= value << 46;
3792 *(target+1) |= value >> 18;
3795 *(target+1) |= value << 23;
3801 ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value)
3803 Elf64_Xword instruction;
3804 Elf64_Sxword rel_value;
3806 rel_value = value - gp_val;
3807 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff))
3808 barf("GP-relative data out of range (address = 0x%lx, gp = 0x%lx)", value, gp_val);
3810 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3811 instruction |= (((rel_value >> 0) & 0x07f) << 13) /* imm7b */
3812 | (((rel_value >> 7) & 0x1ff) << 27) /* imm9d */
3813 | (((rel_value >> 16) & 0x01f) << 22) /* imm5c */
3814 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3815 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3819 ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc)
3821 Elf64_Xword instruction;
3822 Elf64_Sxword rel_value;
3825 entry = allocatePLTEntry(value, oc);
3827 rel_value = (entry >> 4) - (target >> 4);
3828 if ((rel_value > 0xfffff) || (rel_value < -0xfffff))
3829 barf("PLT entry too far away (entry = 0x%lx, target = 0x%lx)", entry, target);
3831 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3832 instruction |= ((rel_value & 0xfffff) << 13) /* imm20b */
3833 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3834 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3840 * PowerPC & X86_64 ELF specifics
3843 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3845 static int ocAllocateSymbolExtras_ELF( ObjectCode *oc )
3851 ehdr = (Elf_Ehdr *) oc->image;
3852 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3854 for( i = 0; i < ehdr->e_shnum; i++ )
3855 if( shdr[i].sh_type == SHT_SYMTAB )
3858 if( i == ehdr->e_shnum )
3860 errorBelch( "This ELF file contains no symtab" );
3864 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3866 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3867 (int) shdr[i].sh_entsize, (int) sizeof( Elf_Sym ) );
3872 return ocAllocateSymbolExtras( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3875 #endif /* powerpc */
3879 /* --------------------------------------------------------------------------
3881 * ------------------------------------------------------------------------*/
3883 #if defined(OBJFORMAT_MACHO)
3886 Support for MachO linking on Darwin/MacOS X
3887 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3889 I hereby formally apologize for the hackish nature of this code.
3890 Things that need to be done:
3891 *) implement ocVerifyImage_MachO
3892 *) add still more sanity checks.
3895 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
3896 #define mach_header mach_header_64
3897 #define segment_command segment_command_64
3898 #define section section_64
3899 #define nlist nlist_64
3902 #ifdef powerpc_HOST_ARCH
3903 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3905 struct mach_header *header = (struct mach_header *) oc->image;
3906 struct load_command *lc = (struct load_command *) (header + 1);
3909 for( i = 0; i < header->ncmds; i++ )
3911 if( lc->cmd == LC_SYMTAB )
3913 // Find out the first and last undefined external
3914 // symbol, so we don't have to allocate too many
3916 struct symtab_command *symLC = (struct symtab_command *) lc;
3917 unsigned min = symLC->nsyms, max = 0;
3918 struct nlist *nlist =
3919 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
3921 for(i=0;i<symLC->nsyms;i++)
3923 if(nlist[i].n_type & N_STAB)
3925 else if(nlist[i].n_type & N_EXT)
3927 if((nlist[i].n_type & N_TYPE) == N_UNDF
3928 && (nlist[i].n_value == 0))
3938 return ocAllocateSymbolExtras(oc, max - min + 1, min);
3943 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3945 return ocAllocateSymbolExtras(oc,0,0);
3948 #ifdef x86_64_HOST_ARCH
3949 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3951 struct mach_header *header = (struct mach_header *) oc->image;
3952 struct load_command *lc = (struct load_command *) (header + 1);
3955 for( i = 0; i < header->ncmds; i++ )
3957 if( lc->cmd == LC_SYMTAB )
3959 // Just allocate one entry for every symbol
3960 struct symtab_command *symLC = (struct symtab_command *) lc;
3962 return ocAllocateSymbolExtras(oc, symLC->nsyms, 0);
3965 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3967 return ocAllocateSymbolExtras(oc,0,0);
3971 static int ocVerifyImage_MachO(ObjectCode* oc)
3973 char *image = (char*) oc->image;
3974 struct mach_header *header = (struct mach_header*) image;
3976 #if x86_64_TARGET_ARCH || powerpc64_TARGET_ARCH
3977 if(header->magic != MH_MAGIC_64)
3980 if(header->magic != MH_MAGIC)
3983 // FIXME: do some more verifying here
3987 static int resolveImports(
3990 struct symtab_command *symLC,
3991 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
3992 unsigned long *indirectSyms,
3993 struct nlist *nlist)
3996 size_t itemSize = 4;
3999 int isJumpTable = 0;
4000 if(!strcmp(sect->sectname,"__jump_table"))
4004 ASSERT(sect->reserved2 == itemSize);
4008 for(i=0; i*itemSize < sect->size;i++)
4010 // according to otool, reserved1 contains the first index into the indirect symbol table
4011 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
4012 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4015 if((symbol->n_type & N_TYPE) == N_UNDF
4016 && (symbol->n_type & N_EXT) && (symbol->n_value != 0))
4017 addr = (void*) (symbol->n_value);
4019 addr = lookupSymbol(nm);
4022 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
4030 checkProddableBlock(oc,image + sect->offset + i*itemSize);
4031 *(image + sect->offset + i*itemSize) = 0xe9; // jmp
4032 *(unsigned*)(image + sect->offset + i*itemSize + 1)
4033 = (char*)addr - (image + sect->offset + i*itemSize + 5);
4038 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
4039 ((void**)(image + sect->offset))[i] = addr;
4046 static unsigned long relocateAddress(
4049 struct section* sections,
4050 unsigned long address)
4053 for(i = 0; i < nSections; i++)
4055 if(sections[i].addr <= address
4056 && address < sections[i].addr + sections[i].size)
4058 return (unsigned long)oc->image
4059 + sections[i].offset + address - sections[i].addr;
4062 barf("Invalid Mach-O file:"
4063 "Address out of bounds while relocating object file");
4067 static int relocateSection(
4070 struct symtab_command *symLC, struct nlist *nlist,
4071 int nSections, struct section* sections, struct section *sect)
4073 struct relocation_info *relocs;
4076 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
4078 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
4080 else if(!strcmp(sect->sectname,"__la_sym_ptr2"))
4082 else if(!strcmp(sect->sectname,"__la_sym_ptr3"))
4086 relocs = (struct relocation_info*) (image + sect->reloff);
4090 #ifdef x86_64_HOST_ARCH
4091 struct relocation_info *reloc = &relocs[i];
4093 char *thingPtr = image + sect->offset + reloc->r_address;
4097 int type = reloc->r_type;
4099 checkProddableBlock(oc,thingPtr);
4100 switch(reloc->r_length)
4103 thing = *(uint8_t*)thingPtr;
4104 baseValue = (uint64_t)thingPtr + 1;
4107 thing = *(uint16_t*)thingPtr;
4108 baseValue = (uint64_t)thingPtr + 2;
4111 thing = *(uint32_t*)thingPtr;
4112 baseValue = (uint64_t)thingPtr + 4;
4115 thing = *(uint64_t*)thingPtr;
4116 baseValue = (uint64_t)thingPtr + 8;
4119 barf("Unknown size.");
4122 if(type == X86_64_RELOC_GOT
4123 || type == X86_64_RELOC_GOT_LOAD)
4125 ASSERT(reloc->r_extern);
4126 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)->addr;
4128 type = X86_64_RELOC_SIGNED;
4130 else if(reloc->r_extern)
4132 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4133 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4134 if(symbol->n_value == 0)
4135 value = (uint64_t) lookupSymbol(nm);
4137 value = relocateAddress(oc, nSections, sections,
4142 value = sections[reloc->r_symbolnum-1].offset
4143 - sections[reloc->r_symbolnum-1].addr
4147 if(type == X86_64_RELOC_BRANCH)
4149 if((int32_t)(value - baseValue) != (int64_t)(value - baseValue))
4151 ASSERT(reloc->r_extern);
4152 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)
4155 ASSERT((int32_t)(value - baseValue) == (int64_t)(value - baseValue));
4156 type = X86_64_RELOC_SIGNED;
4161 case X86_64_RELOC_UNSIGNED:
4162 ASSERT(!reloc->r_pcrel);
4165 case X86_64_RELOC_SIGNED:
4166 ASSERT(reloc->r_pcrel);
4167 thing += value - baseValue;
4169 case X86_64_RELOC_SUBTRACTOR:
4170 ASSERT(!reloc->r_pcrel);
4174 barf("unkown relocation");
4177 switch(reloc->r_length)
4180 *(uint8_t*)thingPtr = thing;
4183 *(uint16_t*)thingPtr = thing;
4186 *(uint32_t*)thingPtr = thing;
4189 *(uint64_t*)thingPtr = thing;
4193 if(relocs[i].r_address & R_SCATTERED)
4195 struct scattered_relocation_info *scat =
4196 (struct scattered_relocation_info*) &relocs[i];
4200 if(scat->r_length == 2)
4202 unsigned long word = 0;
4203 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
4204 checkProddableBlock(oc,wordPtr);
4206 // Note on relocation types:
4207 // i386 uses the GENERIC_RELOC_* types,
4208 // while ppc uses special PPC_RELOC_* types.
4209 // *_RELOC_VANILLA and *_RELOC_PAIR have the same value
4210 // in both cases, all others are different.
4211 // Therefore, we use GENERIC_RELOC_VANILLA
4212 // and GENERIC_RELOC_PAIR instead of the PPC variants,
4213 // and use #ifdefs for the other types.
4215 // Step 1: Figure out what the relocated value should be
4216 if(scat->r_type == GENERIC_RELOC_VANILLA)
4218 word = *wordPtr + (unsigned long) relocateAddress(
4225 #ifdef powerpc_HOST_ARCH
4226 else if(scat->r_type == PPC_RELOC_SECTDIFF
4227 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
4228 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
4229 || scat->r_type == PPC_RELOC_HA16_SECTDIFF)
4231 else if(scat->r_type == GENERIC_RELOC_SECTDIFF)
4234 struct scattered_relocation_info *pair =
4235 (struct scattered_relocation_info*) &relocs[i+1];
4237 if(!pair->r_scattered || pair->r_type != GENERIC_RELOC_PAIR)
4238 barf("Invalid Mach-O file: "
4239 "RELOC_*_SECTDIFF not followed by RELOC_PAIR");
4241 word = (unsigned long)
4242 (relocateAddress(oc, nSections, sections, scat->r_value)
4243 - relocateAddress(oc, nSections, sections, pair->r_value));
4246 #ifdef powerpc_HOST_ARCH
4247 else if(scat->r_type == PPC_RELOC_HI16
4248 || scat->r_type == PPC_RELOC_LO16
4249 || scat->r_type == PPC_RELOC_HA16
4250 || scat->r_type == PPC_RELOC_LO14)
4251 { // these are generated by label+offset things
4252 struct relocation_info *pair = &relocs[i+1];
4253 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
4254 barf("Invalid Mach-O file: "
4255 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
4257 if(scat->r_type == PPC_RELOC_LO16)
4259 word = ((unsigned short*) wordPtr)[1];
4260 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4262 else if(scat->r_type == PPC_RELOC_LO14)
4264 barf("Unsupported Relocation: PPC_RELOC_LO14");
4265 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
4266 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4268 else if(scat->r_type == PPC_RELOC_HI16)
4270 word = ((unsigned short*) wordPtr)[1] << 16;
4271 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4273 else if(scat->r_type == PPC_RELOC_HA16)
4275 word = ((unsigned short*) wordPtr)[1] << 16;
4276 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4280 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
4287 continue; // ignore the others
4289 #ifdef powerpc_HOST_ARCH
4290 if(scat->r_type == GENERIC_RELOC_VANILLA
4291 || scat->r_type == PPC_RELOC_SECTDIFF)
4293 if(scat->r_type == GENERIC_RELOC_VANILLA
4294 || scat->r_type == GENERIC_RELOC_SECTDIFF)
4299 #ifdef powerpc_HOST_ARCH
4300 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
4302 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4304 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
4306 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4308 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
4310 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4311 + ((word & (1<<15)) ? 1 : 0);
4317 continue; // FIXME: I hope it's OK to ignore all the others.
4321 struct relocation_info *reloc = &relocs[i];
4322 if(reloc->r_pcrel && !reloc->r_extern)
4325 if(reloc->r_length == 2)
4327 unsigned long word = 0;
4328 #ifdef powerpc_HOST_ARCH
4329 unsigned long jumpIsland = 0;
4330 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
4331 // to avoid warning and to catch
4335 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
4336 checkProddableBlock(oc,wordPtr);
4338 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4342 #ifdef powerpc_HOST_ARCH
4343 else if(reloc->r_type == PPC_RELOC_LO16)
4345 word = ((unsigned short*) wordPtr)[1];
4346 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4348 else if(reloc->r_type == PPC_RELOC_HI16)
4350 word = ((unsigned short*) wordPtr)[1] << 16;
4351 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4353 else if(reloc->r_type == PPC_RELOC_HA16)
4355 word = ((unsigned short*) wordPtr)[1] << 16;
4356 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4358 else if(reloc->r_type == PPC_RELOC_BR24)
4361 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
4365 if(!reloc->r_extern)
4368 sections[reloc->r_symbolnum-1].offset
4369 - sections[reloc->r_symbolnum-1].addr
4376 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4377 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4378 void *symbolAddress = lookupSymbol(nm);
4381 errorBelch("\nunknown symbol `%s'", nm);
4387 #ifdef powerpc_HOST_ARCH
4388 // In the .o file, this should be a relative jump to NULL
4389 // and we'll change it to a relative jump to the symbol
4390 ASSERT(word + reloc->r_address == 0);
4391 jumpIsland = (unsigned long)
4392 &makeSymbolExtra(oc,
4394 (unsigned long) symbolAddress)
4398 offsetToJumpIsland = word + jumpIsland
4399 - (((long)image) + sect->offset - sect->addr);
4402 word += (unsigned long) symbolAddress
4403 - (((long)image) + sect->offset - sect->addr);
4407 word += (unsigned long) symbolAddress;
4411 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4416 #ifdef powerpc_HOST_ARCH
4417 else if(reloc->r_type == PPC_RELOC_LO16)
4419 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4422 else if(reloc->r_type == PPC_RELOC_HI16)
4424 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4427 else if(reloc->r_type == PPC_RELOC_HA16)
4429 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4430 + ((word & (1<<15)) ? 1 : 0);
4433 else if(reloc->r_type == PPC_RELOC_BR24)
4435 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4437 // The branch offset is too large.
4438 // Therefore, we try to use a jump island.
4441 barf("unconditional relative branch out of range: "
4442 "no jump island available");
4445 word = offsetToJumpIsland;
4446 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4447 barf("unconditional relative branch out of range: "
4448 "jump island out of range");
4450 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
4455 barf("\nunknown relocation %d",reloc->r_type);
4463 static int ocGetNames_MachO(ObjectCode* oc)
4465 char *image = (char*) oc->image;
4466 struct mach_header *header = (struct mach_header*) image;
4467 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4468 unsigned i,curSymbol = 0;
4469 struct segment_command *segLC = NULL;
4470 struct section *sections;
4471 struct symtab_command *symLC = NULL;
4472 struct nlist *nlist;
4473 unsigned long commonSize = 0;
4474 char *commonStorage = NULL;
4475 unsigned long commonCounter;
4477 for(i=0;i<header->ncmds;i++)
4479 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4480 segLC = (struct segment_command*) lc;
4481 else if(lc->cmd == LC_SYMTAB)
4482 symLC = (struct symtab_command*) lc;
4483 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4486 sections = (struct section*) (segLC+1);
4487 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4491 barf("ocGetNames_MachO: no segment load command");
4493 for(i=0;i<segLC->nsects;i++)
4495 if(sections[i].size == 0)
4498 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
4500 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
4501 "ocGetNames_MachO(common symbols)");
4502 sections[i].offset = zeroFillArea - image;
4505 if(!strcmp(sections[i].sectname,"__text"))
4506 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
4507 (void*) (image + sections[i].offset),
4508 (void*) (image + sections[i].offset + sections[i].size));
4509 else if(!strcmp(sections[i].sectname,"__const"))
4510 addSection(oc, SECTIONKIND_RWDATA,
4511 (void*) (image + sections[i].offset),
4512 (void*) (image + sections[i].offset + sections[i].size));
4513 else if(!strcmp(sections[i].sectname,"__data"))
4514 addSection(oc, SECTIONKIND_RWDATA,
4515 (void*) (image + sections[i].offset),
4516 (void*) (image + sections[i].offset + sections[i].size));
4517 else if(!strcmp(sections[i].sectname,"__bss")
4518 || !strcmp(sections[i].sectname,"__common"))
4519 addSection(oc, SECTIONKIND_RWDATA,
4520 (void*) (image + sections[i].offset),
4521 (void*) (image + sections[i].offset + sections[i].size));
4523 addProddableBlock(oc, (void*) (image + sections[i].offset),
4527 // count external symbols defined here
4531 for(i=0;i<symLC->nsyms;i++)
4533 if(nlist[i].n_type & N_STAB)
4535 else if(nlist[i].n_type & N_EXT)
4537 if((nlist[i].n_type & N_TYPE) == N_UNDF
4538 && (nlist[i].n_value != 0))
4540 commonSize += nlist[i].n_value;
4543 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4548 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
4549 "ocGetNames_MachO(oc->symbols)");
4553 for(i=0;i<symLC->nsyms;i++)
4555 if(nlist[i].n_type & N_STAB)
4557 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4559 if(nlist[i].n_type & N_EXT)
4561 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4562 if((nlist[i].n_desc & N_WEAK_DEF) && lookupSymbol(nm))
4563 ; // weak definition, and we already have a definition
4566 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4568 + sections[nlist[i].n_sect-1].offset
4569 - sections[nlist[i].n_sect-1].addr
4570 + nlist[i].n_value);
4571 oc->symbols[curSymbol++] = nm;
4578 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
4579 commonCounter = (unsigned long)commonStorage;
4582 for(i=0;i<symLC->nsyms;i++)
4584 if((nlist[i].n_type & N_TYPE) == N_UNDF
4585 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
4587 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4588 unsigned long sz = nlist[i].n_value;
4590 nlist[i].n_value = commonCounter;
4592 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4593 (void*)commonCounter);
4594 oc->symbols[curSymbol++] = nm;
4596 commonCounter += sz;
4603 static int ocResolve_MachO(ObjectCode* oc)
4605 char *image = (char*) oc->image;
4606 struct mach_header *header = (struct mach_header*) image;
4607 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4609 struct segment_command *segLC = NULL;
4610 struct section *sections;
4611 struct symtab_command *symLC = NULL;
4612 struct dysymtab_command *dsymLC = NULL;
4613 struct nlist *nlist;
4615 for(i=0;i<header->ncmds;i++)
4617 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4618 segLC = (struct segment_command*) lc;
4619 else if(lc->cmd == LC_SYMTAB)
4620 symLC = (struct symtab_command*) lc;
4621 else if(lc->cmd == LC_DYSYMTAB)
4622 dsymLC = (struct dysymtab_command*) lc;
4623 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4626 sections = (struct section*) (segLC+1);
4627 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4632 unsigned long *indirectSyms
4633 = (unsigned long*) (image + dsymLC->indirectsymoff);
4635 for(i=0;i<segLC->nsects;i++)
4637 if( !strcmp(sections[i].sectname,"__la_symbol_ptr")
4638 || !strcmp(sections[i].sectname,"__la_sym_ptr2")
4639 || !strcmp(sections[i].sectname,"__la_sym_ptr3"))
4641 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4644 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr")
4645 || !strcmp(sections[i].sectname,"__pointers"))
4647 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4650 else if(!strcmp(sections[i].sectname,"__jump_table"))
4652 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4658 for(i=0;i<segLC->nsects;i++)
4660 if(!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
4664 #if defined (powerpc_HOST_ARCH)
4665 ocFlushInstructionCache( oc );
4671 #ifdef powerpc_HOST_ARCH
4673 * The Mach-O object format uses leading underscores. But not everywhere.
4674 * There is a small number of runtime support functions defined in
4675 * libcc_dynamic.a whose name does not have a leading underscore.
4676 * As a consequence, we can't get their address from C code.
4677 * We have to use inline assembler just to take the address of a function.
4681 static void machoInitSymbolsWithoutUnderscore()
4683 extern void* symbolsWithoutUnderscore[];
4684 void **p = symbolsWithoutUnderscore;
4685 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
4687 #undef SymI_NeedsProto
4688 #define SymI_NeedsProto(x) \
4689 __asm__ volatile(".long " # x);
4691 RTS_MACHO_NOUNDERLINE_SYMBOLS
4693 __asm__ volatile(".text");
4695 #undef SymI_NeedsProto
4696 #define SymI_NeedsProto(x) \
4697 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
4699 RTS_MACHO_NOUNDERLINE_SYMBOLS
4701 #undef SymI_NeedsProto
4706 * Figure out by how much to shift the entire Mach-O file in memory
4707 * when loading so that its single segment ends up 16-byte-aligned
4709 static int machoGetMisalignment( FILE * f )
4711 struct mach_header header;
4714 fread(&header, sizeof(header), 1, f);
4717 #if x86_64_TARGET_ARCH || powerpc64_TARGET_ARCH
4718 if(header.magic != MH_MAGIC_64)
4721 if(header.magic != MH_MAGIC)
4725 misalignment = (header.sizeofcmds + sizeof(header))
4728 return misalignment ? (16 - misalignment) : 0;