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;
1306 pagesize = getpagesize();
1307 size = ROUND_UP(bytes, pagesize);
1309 #if defined(x86_64_HOST_ARCH)
1312 if (mmap_32bit_base != 0) {
1313 map_addr = mmap_32bit_base;
1317 result = mmap(map_addr, size, PROT_EXEC|PROT_READ|PROT_WRITE,
1318 MAP_PRIVATE|TRY_MAP_32BIT|flags, fd, 0);
1320 if (result == MAP_FAILED) {
1321 sysErrorBelch("mmap");
1322 stg_exit(EXIT_FAILURE);
1325 #if defined(x86_64_HOST_ARCH)
1326 if (mmap_32bit_base != 0) {
1327 if (result == map_addr) {
1328 mmap_32bit_base = map_addr + size;
1330 if ((W_)result > 0x80000000) {
1331 // oops, we were given memory over 2Gb
1332 // ... try allocating memory somewhere else?;
1333 barf("loadObj: failed to mmap() memory below 2Gb; asked for %lu bytes at %p, got %p. Try specifying an address with +RTS -xm<addr> -RTS", size, map_addr, result);
1335 // hmm, we were given memory somewhere else, but it's
1336 // still under 2Gb so we can use it. Next time, ask
1337 // for memory right after the place we just got some
1338 mmap_32bit_base = (void*)result + size;
1342 if ((W_)result > 0x80000000) {
1343 // oops, we were given memory over 2Gb
1344 // ... try allocating memory somewhere else?;
1345 debugTrace(DEBUG_linker,"MAP_32BIT didn't work; gave us %lu bytes at 0x%p", bytes, result);
1346 munmap(result, size);
1348 // Set a base address and try again... (guess: 1Gb)
1349 mmap_32bit_base = (void*)0x40000000;
1359 /* -----------------------------------------------------------------------------
1360 * Load an obj (populate the global symbol table, but don't resolve yet)
1362 * Returns: 1 if ok, 0 on error.
1365 loadObj( char *path )
1377 /* debugBelch("loadObj %s\n", path ); */
1379 /* Check that we haven't already loaded this object.
1380 Ignore requests to load multiple times */
1384 for (o = objects; o; o = o->next) {
1385 if (0 == strcmp(o->fileName, path)) {
1387 break; /* don't need to search further */
1391 IF_DEBUG(linker, debugBelch(
1392 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1393 "same object file twice:\n"
1395 "GHCi will ignore this, but be warned.\n"
1397 return 1; /* success */
1401 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
1403 # if defined(OBJFORMAT_ELF)
1404 oc->formatName = "ELF";
1405 # elif defined(OBJFORMAT_PEi386)
1406 oc->formatName = "PEi386";
1407 # elif defined(OBJFORMAT_MACHO)
1408 oc->formatName = "Mach-O";
1411 barf("loadObj: not implemented on this platform");
1414 r = stat(path, &st);
1415 if (r == -1) { return 0; }
1417 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1418 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1419 strcpy(oc->fileName, path);
1421 oc->fileSize = st.st_size;
1423 oc->sections = NULL;
1424 oc->proddables = NULL;
1426 /* chain it onto the list of objects */
1431 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1433 #if defined(openbsd_HOST_OS)
1434 fd = open(path, O_RDONLY, S_IRUSR);
1436 fd = open(path, O_RDONLY);
1439 barf("loadObj: can't open `%s'", path);
1441 #ifdef ia64_HOST_ARCH
1442 /* The PLT needs to be right before the object */
1445 pagesize = getpagesize();
1446 n = ROUND_UP(PLTSize(), pagesize);
1447 oc->plt = mmap(NULL, n, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
1448 if (oc->plt == MAP_FAILED)
1449 barf("loadObj: can't allocate PLT");
1452 map_addr = oc->plt + n;
1454 n = ROUND_UP(oc->fileSize, pagesize);
1455 oc->image = mmap(map_addr, n, PROT_EXEC|PROT_READ|PROT_WRITE,
1456 MAP_PRIVATE|TRY_MAP_32BIT, fd, 0);
1457 if (oc->image == MAP_FAILED)
1458 barf("loadObj: can't map `%s'", path);
1461 oc->image = mmapForLinker(oc->fileSize, 0, fd);
1466 #else /* !USE_MMAP */
1467 /* load the image into memory */
1468 f = fopen(path, "rb");
1470 barf("loadObj: can't read `%s'", path);
1472 # if defined(mingw32_HOST_OS)
1473 // TODO: We would like to use allocateExec here, but allocateExec
1474 // cannot currently allocate blocks large enough.
1475 oc->image = VirtualAlloc(NULL, oc->fileSize, MEM_RESERVE | MEM_COMMIT,
1476 PAGE_EXECUTE_READWRITE);
1477 # elif defined(darwin_HOST_OS)
1478 // In a Mach-O .o file, all sections can and will be misaligned
1479 // if the total size of the headers is not a multiple of the
1480 // desired alignment. This is fine for .o files that only serve
1481 // as input for the static linker, but it's not fine for us,
1482 // as SSE (used by gcc for floating point) and Altivec require
1483 // 16-byte alignment.
1484 // We calculate the correct alignment from the header before
1485 // reading the file, and then we misalign oc->image on purpose so
1486 // that the actual sections end up aligned again.
1487 oc->misalignment = machoGetMisalignment(f);
1488 oc->image = stgMallocBytes(oc->fileSize + oc->misalignment, "loadObj(image)");
1489 oc->image += oc->misalignment;
1491 oc->image = stgMallocBytes(oc->fileSize, "loadObj(image)");
1496 n = fread ( oc->image, 1, oc->fileSize, f );
1497 if (n != oc->fileSize)
1498 barf("loadObj: error whilst reading `%s'", path);
1501 #endif /* USE_MMAP */
1503 # if defined(OBJFORMAT_MACHO) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1504 r = ocAllocateSymbolExtras_MachO ( oc );
1505 if (!r) { return r; }
1506 # elif defined(OBJFORMAT_ELF) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1507 r = ocAllocateSymbolExtras_ELF ( oc );
1508 if (!r) { return r; }
1511 /* verify the in-memory image */
1512 # if defined(OBJFORMAT_ELF)
1513 r = ocVerifyImage_ELF ( oc );
1514 # elif defined(OBJFORMAT_PEi386)
1515 r = ocVerifyImage_PEi386 ( oc );
1516 # elif defined(OBJFORMAT_MACHO)
1517 r = ocVerifyImage_MachO ( oc );
1519 barf("loadObj: no verify method");
1521 if (!r) { return r; }
1523 /* build the symbol list for this image */
1524 # if defined(OBJFORMAT_ELF)
1525 r = ocGetNames_ELF ( oc );
1526 # elif defined(OBJFORMAT_PEi386)
1527 r = ocGetNames_PEi386 ( oc );
1528 # elif defined(OBJFORMAT_MACHO)
1529 r = ocGetNames_MachO ( oc );
1531 barf("loadObj: no getNames method");
1533 if (!r) { return r; }
1535 /* loaded, but not resolved yet */
1536 oc->status = OBJECT_LOADED;
1541 /* -----------------------------------------------------------------------------
1542 * resolve all the currently unlinked objects in memory
1544 * Returns: 1 if ok, 0 on error.
1554 for (oc = objects; oc; oc = oc->next) {
1555 if (oc->status != OBJECT_RESOLVED) {
1556 # if defined(OBJFORMAT_ELF)
1557 r = ocResolve_ELF ( oc );
1558 # elif defined(OBJFORMAT_PEi386)
1559 r = ocResolve_PEi386 ( oc );
1560 # elif defined(OBJFORMAT_MACHO)
1561 r = ocResolve_MachO ( oc );
1563 barf("resolveObjs: not implemented on this platform");
1565 if (!r) { return r; }
1566 oc->status = OBJECT_RESOLVED;
1572 /* -----------------------------------------------------------------------------
1573 * delete an object from the pool
1576 unloadObj( char *path )
1578 ObjectCode *oc, *prev;
1580 ASSERT(symhash != NULL);
1581 ASSERT(objects != NULL);
1586 for (oc = objects; oc; prev = oc, oc = oc->next) {
1587 if (!strcmp(oc->fileName,path)) {
1589 /* Remove all the mappings for the symbols within this
1594 for (i = 0; i < oc->n_symbols; i++) {
1595 if (oc->symbols[i] != NULL) {
1596 removeStrHashTable(symhash, oc->symbols[i], NULL);
1604 prev->next = oc->next;
1607 // We're going to leave this in place, in case there are
1608 // any pointers from the heap into it:
1609 // #ifdef mingw32_HOST_OS
1610 // VirtualFree(oc->image);
1612 // stgFree(oc->image);
1614 stgFree(oc->fileName);
1615 stgFree(oc->symbols);
1616 stgFree(oc->sections);
1622 errorBelch("unloadObj: can't find `%s' to unload", path);
1626 /* -----------------------------------------------------------------------------
1627 * Sanity checking. For each ObjectCode, maintain a list of address ranges
1628 * which may be prodded during relocation, and abort if we try and write
1629 * outside any of these.
1631 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
1634 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
1635 /* debugBelch("aPB %p %p %d\n", oc, start, size); */
1639 pb->next = oc->proddables;
1640 oc->proddables = pb;
1643 static void checkProddableBlock ( ObjectCode* oc, void* addr )
1646 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
1647 char* s = (char*)(pb->start);
1648 char* e = s + pb->size - 1;
1649 char* a = (char*)addr;
1650 /* Assumes that the biggest fixup involves a 4-byte write. This
1651 probably needs to be changed to 8 (ie, +7) on 64-bit
1653 if (a >= s && (a+3) <= e) return;
1655 barf("checkProddableBlock: invalid fixup in runtime linker");
1658 /* -----------------------------------------------------------------------------
1659 * Section management.
1661 static void addSection ( ObjectCode* oc, SectionKind kind,
1662 void* start, void* end )
1664 Section* s = stgMallocBytes(sizeof(Section), "addSection");
1668 s->next = oc->sections;
1671 debugBelch("addSection: %p-%p (size %d), kind %d\n",
1672 start, ((char*)end)-1, end - start + 1, kind );
1677 /* --------------------------------------------------------------------------
1679 * This is about allocating a small chunk of memory for every symbol in the
1680 * object file. We make sure that the SymboLExtras are always "in range" of
1681 * limited-range PC-relative instructions on various platforms by allocating
1682 * them right next to the object code itself.
1685 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
1688 ocAllocateSymbolExtras
1690 Allocate additional space at the end of the object file image to make room
1691 for jump islands (powerpc, x86_64) and GOT entries (x86_64).
1693 PowerPC relative branch instructions have a 24 bit displacement field.
1694 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
1695 If a particular imported symbol is outside this range, we have to redirect
1696 the jump to a short piece of new code that just loads the 32bit absolute
1697 address and jumps there.
1698 On x86_64, PC-relative jumps and PC-relative accesses to the GOT are limited
1701 This function just allocates space for one SymbolExtra for every
1702 undefined symbol in the object file. The code for the jump islands is
1703 filled in by makeSymbolExtra below.
1706 static int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first )
1713 int misalignment = 0;
1714 #ifdef darwin_HOST_OS
1715 misalignment = oc->misalignment;
1721 // round up to the nearest 4
1722 aligned = (oc->fileSize + 3) & ~3;
1725 pagesize = getpagesize();
1726 n = ROUND_UP( oc->fileSize, pagesize );
1727 m = ROUND_UP( aligned + sizeof (SymbolExtra) * count, pagesize );
1729 /* we try to use spare space at the end of the last page of the
1730 * image for the jump islands, but if there isn't enough space
1731 * then we have to map some (anonymously, remembering MAP_32BIT).
1733 if( m > n ) // we need to allocate more pages
1735 oc->symbol_extras = mmapForLinker(sizeof(SymbolExtra) * count,
1740 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1743 oc->image -= misalignment;
1744 oc->image = stgReallocBytes( oc->image,
1746 aligned + sizeof (SymbolExtra) * count,
1747 "ocAllocateSymbolExtras" );
1748 oc->image += misalignment;
1750 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1751 #endif /* USE_MMAP */
1753 memset( oc->symbol_extras, 0, sizeof (SymbolExtra) * count );
1756 oc->symbol_extras = NULL;
1758 oc->first_symbol_extra = first;
1759 oc->n_symbol_extras = count;
1764 static SymbolExtra* makeSymbolExtra( ObjectCode* oc,
1765 unsigned long symbolNumber,
1766 unsigned long target )
1770 ASSERT( symbolNumber >= oc->first_symbol_extra
1771 && symbolNumber - oc->first_symbol_extra < oc->n_symbol_extras);
1773 extra = &oc->symbol_extras[symbolNumber - oc->first_symbol_extra];
1775 #ifdef powerpc_HOST_ARCH
1776 // lis r12, hi16(target)
1777 extra->jumpIsland.lis_r12 = 0x3d80;
1778 extra->jumpIsland.hi_addr = target >> 16;
1780 // ori r12, r12, lo16(target)
1781 extra->jumpIsland.ori_r12_r12 = 0x618c;
1782 extra->jumpIsland.lo_addr = target & 0xffff;
1785 extra->jumpIsland.mtctr_r12 = 0x7d8903a6;
1788 extra->jumpIsland.bctr = 0x4e800420;
1790 #ifdef x86_64_HOST_ARCH
1792 static uint8_t jmp[] = { 0xFF, 0x25, 0xF2, 0xFF, 0xFF, 0xFF };
1793 extra->addr = target;
1794 memcpy(extra->jumpIsland, jmp, 6);
1802 /* --------------------------------------------------------------------------
1803 * PowerPC specifics (instruction cache flushing)
1804 * ------------------------------------------------------------------------*/
1806 #ifdef powerpc_TARGET_ARCH
1808 ocFlushInstructionCache
1810 Flush the data & instruction caches.
1811 Because the PPC has split data/instruction caches, we have to
1812 do that whenever we modify code at runtime.
1815 static void ocFlushInstructionCache( ObjectCode *oc )
1817 int n = (oc->fileSize + sizeof( SymbolExtra ) * oc->n_symbol_extras + 3) / 4;
1818 unsigned long *p = (unsigned long *) oc->image;
1822 __asm__ volatile ( "dcbf 0,%0\n\t"
1830 __asm__ volatile ( "sync\n\t"
1836 /* --------------------------------------------------------------------------
1837 * PEi386 specifics (Win32 targets)
1838 * ------------------------------------------------------------------------*/
1840 /* The information for this linker comes from
1841 Microsoft Portable Executable
1842 and Common Object File Format Specification
1843 revision 5.1 January 1998
1844 which SimonM says comes from the MS Developer Network CDs.
1846 It can be found there (on older CDs), but can also be found
1849 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
1851 (this is Rev 6.0 from February 1999).
1853 Things move, so if that fails, try searching for it via
1855 http://www.google.com/search?q=PE+COFF+specification
1857 The ultimate reference for the PE format is the Winnt.h
1858 header file that comes with the Platform SDKs; as always,
1859 implementations will drift wrt their documentation.
1861 A good background article on the PE format is Matt Pietrek's
1862 March 1994 article in Microsoft System Journal (MSJ)
1863 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
1864 Win32 Portable Executable File Format." The info in there
1865 has recently been updated in a two part article in
1866 MSDN magazine, issues Feb and March 2002,
1867 "Inside Windows: An In-Depth Look into the Win32 Portable
1868 Executable File Format"
1870 John Levine's book "Linkers and Loaders" contains useful
1875 #if defined(OBJFORMAT_PEi386)
1879 typedef unsigned char UChar;
1880 typedef unsigned short UInt16;
1881 typedef unsigned int UInt32;
1888 UInt16 NumberOfSections;
1889 UInt32 TimeDateStamp;
1890 UInt32 PointerToSymbolTable;
1891 UInt32 NumberOfSymbols;
1892 UInt16 SizeOfOptionalHeader;
1893 UInt16 Characteristics;
1897 #define sizeof_COFF_header 20
1904 UInt32 VirtualAddress;
1905 UInt32 SizeOfRawData;
1906 UInt32 PointerToRawData;
1907 UInt32 PointerToRelocations;
1908 UInt32 PointerToLinenumbers;
1909 UInt16 NumberOfRelocations;
1910 UInt16 NumberOfLineNumbers;
1911 UInt32 Characteristics;
1915 #define sizeof_COFF_section 40
1922 UInt16 SectionNumber;
1925 UChar NumberOfAuxSymbols;
1929 #define sizeof_COFF_symbol 18
1934 UInt32 VirtualAddress;
1935 UInt32 SymbolTableIndex;
1940 #define sizeof_COFF_reloc 10
1943 /* From PE spec doc, section 3.3.2 */
1944 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
1945 windows.h -- for the same purpose, but I want to know what I'm
1947 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
1948 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
1949 #define MYIMAGE_FILE_DLL 0x2000
1950 #define MYIMAGE_FILE_SYSTEM 0x1000
1951 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
1952 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
1953 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
1955 /* From PE spec doc, section 5.4.2 and 5.4.4 */
1956 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
1957 #define MYIMAGE_SYM_CLASS_STATIC 3
1958 #define MYIMAGE_SYM_UNDEFINED 0
1960 /* From PE spec doc, section 4.1 */
1961 #define MYIMAGE_SCN_CNT_CODE 0x00000020
1962 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
1963 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
1965 /* From PE spec doc, section 5.2.1 */
1966 #define MYIMAGE_REL_I386_DIR32 0x0006
1967 #define MYIMAGE_REL_I386_REL32 0x0014
1970 /* We use myindex to calculate array addresses, rather than
1971 simply doing the normal subscript thing. That's because
1972 some of the above structs have sizes which are not
1973 a whole number of words. GCC rounds their sizes up to a
1974 whole number of words, which means that the address calcs
1975 arising from using normal C indexing or pointer arithmetic
1976 are just plain wrong. Sigh.
1979 myindex ( int scale, void* base, int index )
1982 ((UChar*)base) + scale * index;
1987 printName ( UChar* name, UChar* strtab )
1989 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1990 UInt32 strtab_offset = * (UInt32*)(name+4);
1991 debugBelch("%s", strtab + strtab_offset );
1994 for (i = 0; i < 8; i++) {
1995 if (name[i] == 0) break;
1996 debugBelch("%c", name[i] );
2003 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
2005 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2006 UInt32 strtab_offset = * (UInt32*)(name+4);
2007 strncpy ( dst, strtab+strtab_offset, dstSize );
2013 if (name[i] == 0) break;
2023 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
2026 /* If the string is longer than 8 bytes, look in the
2027 string table for it -- this will be correctly zero terminated.
2029 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2030 UInt32 strtab_offset = * (UInt32*)(name+4);
2031 return ((UChar*)strtab) + strtab_offset;
2033 /* Otherwise, if shorter than 8 bytes, return the original,
2034 which by defn is correctly terminated.
2036 if (name[7]==0) return name;
2037 /* The annoying case: 8 bytes. Copy into a temporary
2038 (which is never freed ...)
2040 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
2042 strncpy(newstr,name,8);
2048 /* Just compares the short names (first 8 chars) */
2049 static COFF_section *
2050 findPEi386SectionCalled ( ObjectCode* oc, char* name )
2054 = (COFF_header*)(oc->image);
2055 COFF_section* sectab
2057 ((UChar*)(oc->image))
2058 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2060 for (i = 0; i < hdr->NumberOfSections; i++) {
2063 COFF_section* section_i
2065 myindex ( sizeof_COFF_section, sectab, i );
2066 n1 = (UChar*) &(section_i->Name);
2068 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
2069 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
2070 n1[6]==n2[6] && n1[7]==n2[7])
2079 zapTrailingAtSign ( UChar* sym )
2081 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
2083 if (sym[0] == 0) return;
2085 while (sym[i] != 0) i++;
2088 while (j > 0 && my_isdigit(sym[j])) j--;
2089 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
2094 lookupSymbolInDLLs ( UChar *lbl )
2099 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
2100 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
2102 if (lbl[0] == '_') {
2103 /* HACK: if the name has an initial underscore, try stripping
2104 it off & look that up first. I've yet to verify whether there's
2105 a Rule that governs whether an initial '_' *should always* be
2106 stripped off when mapping from import lib name to the DLL name.
2108 sym = GetProcAddress(o_dll->instance, (lbl+1));
2110 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
2114 sym = GetProcAddress(o_dll->instance, lbl);
2116 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
2125 ocVerifyImage_PEi386 ( ObjectCode* oc )
2130 COFF_section* sectab;
2131 COFF_symbol* symtab;
2133 /* debugBelch("\nLOADING %s\n", oc->fileName); */
2134 hdr = (COFF_header*)(oc->image);
2135 sectab = (COFF_section*) (
2136 ((UChar*)(oc->image))
2137 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2139 symtab = (COFF_symbol*) (
2140 ((UChar*)(oc->image))
2141 + hdr->PointerToSymbolTable
2143 strtab = ((UChar*)symtab)
2144 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2146 if (hdr->Machine != 0x14c) {
2147 errorBelch("%s: Not x86 PEi386", oc->fileName);
2150 if (hdr->SizeOfOptionalHeader != 0) {
2151 errorBelch("%s: PEi386 with nonempty optional header", oc->fileName);
2154 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
2155 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
2156 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
2157 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
2158 errorBelch("%s: Not a PEi386 object file", oc->fileName);
2161 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
2162 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
2163 errorBelch("%s: Invalid PEi386 word size or endiannness: %d",
2165 (int)(hdr->Characteristics));
2168 /* If the string table size is way crazy, this might indicate that
2169 there are more than 64k relocations, despite claims to the
2170 contrary. Hence this test. */
2171 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
2173 if ( (*(UInt32*)strtab) > 600000 ) {
2174 /* Note that 600k has no special significance other than being
2175 big enough to handle the almost-2MB-sized lumps that
2176 constitute HSwin32*.o. */
2177 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
2182 /* No further verification after this point; only debug printing. */
2184 IF_DEBUG(linker, i=1);
2185 if (i == 0) return 1;
2187 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
2188 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
2189 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
2192 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
2193 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
2194 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
2195 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
2196 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
2197 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
2198 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
2200 /* Print the section table. */
2202 for (i = 0; i < hdr->NumberOfSections; i++) {
2204 COFF_section* sectab_i
2206 myindex ( sizeof_COFF_section, sectab, i );
2213 printName ( sectab_i->Name, strtab );
2223 sectab_i->VirtualSize,
2224 sectab_i->VirtualAddress,
2225 sectab_i->SizeOfRawData,
2226 sectab_i->PointerToRawData,
2227 sectab_i->NumberOfRelocations,
2228 sectab_i->PointerToRelocations,
2229 sectab_i->PointerToRawData
2231 reltab = (COFF_reloc*) (
2232 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2235 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2236 /* If the relocation field (a short) has overflowed, the
2237 * real count can be found in the first reloc entry.
2239 * See Section 4.1 (last para) of the PE spec (rev6.0).
2241 COFF_reloc* rel = (COFF_reloc*)
2242 myindex ( sizeof_COFF_reloc, reltab, 0 );
2243 noRelocs = rel->VirtualAddress;
2246 noRelocs = sectab_i->NumberOfRelocations;
2250 for (; j < noRelocs; j++) {
2252 COFF_reloc* rel = (COFF_reloc*)
2253 myindex ( sizeof_COFF_reloc, reltab, j );
2255 " type 0x%-4x vaddr 0x%-8x name `",
2257 rel->VirtualAddress );
2258 sym = (COFF_symbol*)
2259 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
2260 /* Hmm..mysterious looking offset - what's it for? SOF */
2261 printName ( sym->Name, strtab -10 );
2268 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
2269 debugBelch("---START of string table---\n");
2270 for (i = 4; i < *(Int32*)strtab; i++) {
2272 debugBelch("\n"); else
2273 debugBelch("%c", strtab[i] );
2275 debugBelch("--- END of string table---\n");
2280 COFF_symbol* symtab_i;
2281 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2282 symtab_i = (COFF_symbol*)
2283 myindex ( sizeof_COFF_symbol, symtab, i );
2289 printName ( symtab_i->Name, strtab );
2298 (Int32)(symtab_i->SectionNumber),
2299 (UInt32)symtab_i->Type,
2300 (UInt32)symtab_i->StorageClass,
2301 (UInt32)symtab_i->NumberOfAuxSymbols
2303 i += symtab_i->NumberOfAuxSymbols;
2313 ocGetNames_PEi386 ( ObjectCode* oc )
2316 COFF_section* sectab;
2317 COFF_symbol* symtab;
2324 hdr = (COFF_header*)(oc->image);
2325 sectab = (COFF_section*) (
2326 ((UChar*)(oc->image))
2327 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2329 symtab = (COFF_symbol*) (
2330 ((UChar*)(oc->image))
2331 + hdr->PointerToSymbolTable
2333 strtab = ((UChar*)(oc->image))
2334 + hdr->PointerToSymbolTable
2335 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2337 /* Allocate space for any (local, anonymous) .bss sections. */
2339 for (i = 0; i < hdr->NumberOfSections; i++) {
2342 COFF_section* sectab_i
2344 myindex ( sizeof_COFF_section, sectab, i );
2345 if (0 != strcmp(sectab_i->Name, ".bss")) continue;
2346 /* sof 10/05: the PE spec text isn't too clear regarding what
2347 * the SizeOfRawData field is supposed to hold for object
2348 * file sections containing just uninitialized data -- for executables,
2349 * it is supposed to be zero; unclear what it's supposed to be
2350 * for object files. However, VirtualSize is guaranteed to be
2351 * zero for object files, which definitely suggests that SizeOfRawData
2352 * will be non-zero (where else would the size of this .bss section be
2353 * stored?) Looking at the COFF_section info for incoming object files,
2354 * this certainly appears to be the case.
2356 * => I suspect we've been incorrectly handling .bss sections in (relocatable)
2357 * object files up until now. This turned out to bite us with ghc-6.4.1's use
2358 * of gcc-3.4.x, which has started to emit initially-zeroed-out local 'static'
2359 * variable decls into to the .bss section. (The specific function in Q which
2360 * triggered this is libraries/base/cbits/dirUtils.c:__hscore_getFolderPath())
2362 if (sectab_i->VirtualSize == 0 && sectab_i->SizeOfRawData == 0) continue;
2363 /* This is a non-empty .bss section. Allocate zeroed space for
2364 it, and set its PointerToRawData field such that oc->image +
2365 PointerToRawData == addr_of_zeroed_space. */
2366 bss_sz = sectab_i->VirtualSize;
2367 if ( bss_sz < sectab_i->SizeOfRawData) { bss_sz = sectab_i->SizeOfRawData; }
2368 zspace = stgCallocBytes(1, bss_sz, "ocGetNames_PEi386(anonymous bss)");
2369 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
2370 addProddableBlock(oc, zspace, bss_sz);
2371 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
2374 /* Copy section information into the ObjectCode. */
2376 for (i = 0; i < hdr->NumberOfSections; i++) {
2382 = SECTIONKIND_OTHER;
2383 COFF_section* sectab_i
2385 myindex ( sizeof_COFF_section, sectab, i );
2386 IF_DEBUG(linker, debugBelch("section name = %s\n", sectab_i->Name ));
2389 /* I'm sure this is the Right Way to do it. However, the
2390 alternative of testing the sectab_i->Name field seems to
2391 work ok with Cygwin.
2393 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
2394 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
2395 kind = SECTIONKIND_CODE_OR_RODATA;
2398 if (0==strcmp(".text",sectab_i->Name) ||
2399 0==strcmp(".rdata",sectab_i->Name)||
2400 0==strcmp(".rodata",sectab_i->Name))
2401 kind = SECTIONKIND_CODE_OR_RODATA;
2402 if (0==strcmp(".data",sectab_i->Name) ||
2403 0==strcmp(".bss",sectab_i->Name))
2404 kind = SECTIONKIND_RWDATA;
2406 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
2407 sz = sectab_i->SizeOfRawData;
2408 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
2410 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
2411 end = start + sz - 1;
2413 if (kind == SECTIONKIND_OTHER
2414 /* Ignore sections called which contain stabs debugging
2416 && 0 != strcmp(".stab", sectab_i->Name)
2417 && 0 != strcmp(".stabstr", sectab_i->Name)
2418 /* ignore constructor section for now */
2419 && 0 != strcmp(".ctors", sectab_i->Name)
2420 /* ignore section generated from .ident */
2421 && 0!= strcmp("/4", sectab_i->Name)
2422 /* ignore unknown section that appeared in gcc 3.4.5(?) */
2423 && 0!= strcmp(".reloc", sectab_i->Name)
2425 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName);
2429 if (kind != SECTIONKIND_OTHER && end >= start) {
2430 addSection(oc, kind, start, end);
2431 addProddableBlock(oc, start, end - start + 1);
2435 /* Copy exported symbols into the ObjectCode. */
2437 oc->n_symbols = hdr->NumberOfSymbols;
2438 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2439 "ocGetNames_PEi386(oc->symbols)");
2440 /* Call me paranoid; I don't care. */
2441 for (i = 0; i < oc->n_symbols; i++)
2442 oc->symbols[i] = NULL;
2446 COFF_symbol* symtab_i;
2447 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2448 symtab_i = (COFF_symbol*)
2449 myindex ( sizeof_COFF_symbol, symtab, i );
2453 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2454 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2455 /* This symbol is global and defined, viz, exported */
2456 /* for MYIMAGE_SYMCLASS_EXTERNAL
2457 && !MYIMAGE_SYM_UNDEFINED,
2458 the address of the symbol is:
2459 address of relevant section + offset in section
2461 COFF_section* sectabent
2462 = (COFF_section*) myindex ( sizeof_COFF_section,
2464 symtab_i->SectionNumber-1 );
2465 addr = ((UChar*)(oc->image))
2466 + (sectabent->PointerToRawData
2470 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2471 && symtab_i->Value > 0) {
2472 /* This symbol isn't in any section at all, ie, global bss.
2473 Allocate zeroed space for it. */
2474 addr = stgCallocBytes(1, symtab_i->Value,
2475 "ocGetNames_PEi386(non-anonymous bss)");
2476 addSection(oc, SECTIONKIND_RWDATA, addr,
2477 ((UChar*)addr) + symtab_i->Value - 1);
2478 addProddableBlock(oc, addr, symtab_i->Value);
2479 /* debugBelch("BSS section at 0x%x\n", addr); */
2482 if (addr != NULL ) {
2483 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2484 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2485 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2486 ASSERT(i >= 0 && i < oc->n_symbols);
2487 /* cstring_from_COFF_symbol_name always succeeds. */
2488 oc->symbols[i] = sname;
2489 ghciInsertStrHashTable(oc->fileName, symhash, sname, addr);
2493 "IGNORING symbol %d\n"
2497 printName ( symtab_i->Name, strtab );
2506 (Int32)(symtab_i->SectionNumber),
2507 (UInt32)symtab_i->Type,
2508 (UInt32)symtab_i->StorageClass,
2509 (UInt32)symtab_i->NumberOfAuxSymbols
2514 i += symtab_i->NumberOfAuxSymbols;
2523 ocResolve_PEi386 ( ObjectCode* oc )
2526 COFF_section* sectab;
2527 COFF_symbol* symtab;
2537 /* ToDo: should be variable-sized? But is at least safe in the
2538 sense of buffer-overrun-proof. */
2540 /* debugBelch("resolving for %s\n", oc->fileName); */
2542 hdr = (COFF_header*)(oc->image);
2543 sectab = (COFF_section*) (
2544 ((UChar*)(oc->image))
2545 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2547 symtab = (COFF_symbol*) (
2548 ((UChar*)(oc->image))
2549 + hdr->PointerToSymbolTable
2551 strtab = ((UChar*)(oc->image))
2552 + hdr->PointerToSymbolTable
2553 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2555 for (i = 0; i < hdr->NumberOfSections; i++) {
2556 COFF_section* sectab_i
2558 myindex ( sizeof_COFF_section, sectab, i );
2561 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2564 /* Ignore sections called which contain stabs debugging
2566 if (0 == strcmp(".stab", sectab_i->Name)
2567 || 0 == strcmp(".stabstr", sectab_i->Name)
2568 || 0 == strcmp(".ctors", sectab_i->Name))
2571 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2572 /* If the relocation field (a short) has overflowed, the
2573 * real count can be found in the first reloc entry.
2575 * See Section 4.1 (last para) of the PE spec (rev6.0).
2577 * Nov2003 update: the GNU linker still doesn't correctly
2578 * handle the generation of relocatable object files with
2579 * overflown relocations. Hence the output to warn of potential
2582 COFF_reloc* rel = (COFF_reloc*)
2583 myindex ( sizeof_COFF_reloc, reltab, 0 );
2584 noRelocs = rel->VirtualAddress;
2586 /* 10/05: we now assume (and check for) a GNU ld that is capable
2587 * of handling object files with (>2^16) of relocs.
2590 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2595 noRelocs = sectab_i->NumberOfRelocations;
2600 for (; j < noRelocs; j++) {
2602 COFF_reloc* reltab_j
2604 myindex ( sizeof_COFF_reloc, reltab, j );
2606 /* the location to patch */
2608 ((UChar*)(oc->image))
2609 + (sectab_i->PointerToRawData
2610 + reltab_j->VirtualAddress
2611 - sectab_i->VirtualAddress )
2613 /* the existing contents of pP */
2615 /* the symbol to connect to */
2616 sym = (COFF_symbol*)
2617 myindex ( sizeof_COFF_symbol,
2618 symtab, reltab_j->SymbolTableIndex );
2621 "reloc sec %2d num %3d: type 0x%-4x "
2622 "vaddr 0x%-8x name `",
2624 (UInt32)reltab_j->Type,
2625 reltab_j->VirtualAddress );
2626 printName ( sym->Name, strtab );
2627 debugBelch("'\n" ));
2629 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2630 COFF_section* section_sym
2631 = findPEi386SectionCalled ( oc, sym->Name );
2633 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2636 S = ((UInt32)(oc->image))
2637 + (section_sym->PointerToRawData
2640 copyName ( sym->Name, strtab, symbol, 1000-1 );
2641 S = (UInt32) lookupSymbol( symbol );
2642 if ((void*)S != NULL) goto foundit;
2643 /* Newline first because the interactive linker has printed "linking..." */
2644 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, symbol);
2648 checkProddableBlock(oc, pP);
2649 switch (reltab_j->Type) {
2650 case MYIMAGE_REL_I386_DIR32:
2653 case MYIMAGE_REL_I386_REL32:
2654 /* Tricky. We have to insert a displacement at
2655 pP which, when added to the PC for the _next_
2656 insn, gives the address of the target (S).
2657 Problem is to know the address of the next insn
2658 when we only know pP. We assume that this
2659 literal field is always the last in the insn,
2660 so that the address of the next insn is pP+4
2661 -- hence the constant 4.
2662 Also I don't know if A should be added, but so
2663 far it has always been zero.
2665 SOF 05/2005: 'A' (old contents of *pP) have been observed
2666 to contain values other than zero (the 'wx' object file
2667 that came with wxhaskell-0.9.4; dunno how it was compiled..).
2668 So, add displacement to old value instead of asserting
2669 A to be zero. Fixes wxhaskell-related crashes, and no other
2670 ill effects have been observed.
2672 Update: the reason why we're seeing these more elaborate
2673 relocations is due to a switch in how the NCG compiles SRTs
2674 and offsets to them from info tables. SRTs live in .(ro)data,
2675 while info tables live in .text, causing GAS to emit REL32/DISP32
2676 relocations with non-zero values. Adding the displacement is
2677 the right thing to do.
2679 *pP = S - ((UInt32)pP) - 4 + A;
2682 debugBelch("%s: unhandled PEi386 relocation type %d",
2683 oc->fileName, reltab_j->Type);
2690 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2694 #endif /* defined(OBJFORMAT_PEi386) */
2697 /* --------------------------------------------------------------------------
2699 * ------------------------------------------------------------------------*/
2701 #if defined(OBJFORMAT_ELF)
2706 #if defined(sparc_HOST_ARCH)
2707 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2708 #elif defined(i386_HOST_ARCH)
2709 # define ELF_TARGET_386 /* Used inside <elf.h> */
2710 #elif defined(x86_64_HOST_ARCH)
2711 # define ELF_TARGET_X64_64
2713 #elif defined (ia64_HOST_ARCH)
2714 # define ELF_TARGET_IA64 /* Used inside <elf.h> */
2716 # define ELF_FUNCTION_DESC /* calling convention uses function descriptors */
2717 # define ELF_NEED_GOT /* needs Global Offset Table */
2718 # define ELF_NEED_PLT /* needs Procedure Linkage Tables */
2721 #if !defined(openbsd_HOST_OS)
2724 /* openbsd elf has things in different places, with diff names */
2725 # include <elf_abi.h>
2726 # include <machine/reloc.h>
2727 # define R_386_32 RELOC_32
2728 # define R_386_PC32 RELOC_PC32
2731 /* If elf.h doesn't define it */
2732 # ifndef R_X86_64_PC64
2733 # define R_X86_64_PC64 24
2737 * Define a set of types which can be used for both ELF32 and ELF64
2741 #define ELFCLASS ELFCLASS64
2742 #define Elf_Addr Elf64_Addr
2743 #define Elf_Word Elf64_Word
2744 #define Elf_Sword Elf64_Sword
2745 #define Elf_Ehdr Elf64_Ehdr
2746 #define Elf_Phdr Elf64_Phdr
2747 #define Elf_Shdr Elf64_Shdr
2748 #define Elf_Sym Elf64_Sym
2749 #define Elf_Rel Elf64_Rel
2750 #define Elf_Rela Elf64_Rela
2751 #define ELF_ST_TYPE ELF64_ST_TYPE
2752 #define ELF_ST_BIND ELF64_ST_BIND
2753 #define ELF_R_TYPE ELF64_R_TYPE
2754 #define ELF_R_SYM ELF64_R_SYM
2756 #define ELFCLASS ELFCLASS32
2757 #define Elf_Addr Elf32_Addr
2758 #define Elf_Word Elf32_Word
2759 #define Elf_Sword Elf32_Sword
2760 #define Elf_Ehdr Elf32_Ehdr
2761 #define Elf_Phdr Elf32_Phdr
2762 #define Elf_Shdr Elf32_Shdr
2763 #define Elf_Sym Elf32_Sym
2764 #define Elf_Rel Elf32_Rel
2765 #define Elf_Rela Elf32_Rela
2767 #define ELF_ST_TYPE ELF32_ST_TYPE
2770 #define ELF_ST_BIND ELF32_ST_BIND
2773 #define ELF_R_TYPE ELF32_R_TYPE
2776 #define ELF_R_SYM ELF32_R_SYM
2782 * Functions to allocate entries in dynamic sections. Currently we simply
2783 * preallocate a large number, and we don't check if a entry for the given
2784 * target already exists (a linear search is too slow). Ideally these
2785 * entries would be associated with symbols.
2788 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
2789 #define GOT_SIZE 0x20000
2790 #define FUNCTION_TABLE_SIZE 0x10000
2791 #define PLT_SIZE 0x08000
2794 static Elf_Addr got[GOT_SIZE];
2795 static unsigned int gotIndex;
2796 static Elf_Addr gp_val = (Elf_Addr)got;
2799 allocateGOTEntry(Elf_Addr target)
2803 if (gotIndex >= GOT_SIZE)
2804 barf("Global offset table overflow");
2806 entry = &got[gotIndex++];
2808 return (Elf_Addr)entry;
2812 #ifdef ELF_FUNCTION_DESC
2818 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
2819 static unsigned int functionTableIndex;
2822 allocateFunctionDesc(Elf_Addr target)
2824 FunctionDesc *entry;
2826 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
2827 barf("Function table overflow");
2829 entry = &functionTable[functionTableIndex++];
2831 entry->gp = (Elf_Addr)gp_val;
2832 return (Elf_Addr)entry;
2836 copyFunctionDesc(Elf_Addr target)
2838 FunctionDesc *olddesc = (FunctionDesc *)target;
2839 FunctionDesc *newdesc;
2841 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
2842 newdesc->gp = olddesc->gp;
2843 return (Elf_Addr)newdesc;
2848 #ifdef ia64_HOST_ARCH
2849 static void ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value);
2850 static void ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc);
2852 static unsigned char plt_code[] =
2854 /* taken from binutils bfd/elfxx-ia64.c */
2855 0x0b, 0x78, 0x00, 0x02, 0x00, 0x24, /* [MMI] addl r15=0,r1;; */
2856 0x00, 0x41, 0x3c, 0x30, 0x28, 0xc0, /* ld8 r16=[r15],8 */
2857 0x01, 0x08, 0x00, 0x84, /* mov r14=r1;; */
2858 0x11, 0x08, 0x00, 0x1e, 0x18, 0x10, /* [MIB] ld8 r1=[r15] */
2859 0x60, 0x80, 0x04, 0x80, 0x03, 0x00, /* mov b6=r16 */
2860 0x60, 0x00, 0x80, 0x00 /* br.few b6;; */
2863 /* If we can't get to the function descriptor via gp, take a local copy of it */
2864 #define PLT_RELOC(code, target) { \
2865 Elf64_Sxword rel_value = target - gp_val; \
2866 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff)) \
2867 ia64_reloc_gprel22((Elf_Addr)code, copyFunctionDesc(target)); \
2869 ia64_reloc_gprel22((Elf_Addr)code, target); \
2874 unsigned char code[sizeof(plt_code)];
2878 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
2880 PLTEntry *plt = (PLTEntry *)oc->plt;
2883 if (oc->pltIndex >= PLT_SIZE)
2884 barf("Procedure table overflow");
2886 entry = &plt[oc->pltIndex++];
2887 memcpy(entry->code, plt_code, sizeof(entry->code));
2888 PLT_RELOC(entry->code, target);
2889 return (Elf_Addr)entry;
2895 return (PLT_SIZE * sizeof(PLTEntry));
2901 * Generic ELF functions
2905 findElfSection ( void* objImage, Elf_Word sh_type )
2907 char* ehdrC = (char*)objImage;
2908 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2909 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
2910 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2914 for (i = 0; i < ehdr->e_shnum; i++) {
2915 if (shdr[i].sh_type == sh_type
2916 /* Ignore the section header's string table. */
2917 && i != ehdr->e_shstrndx
2918 /* Ignore string tables named .stabstr, as they contain
2920 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2922 ptr = ehdrC + shdr[i].sh_offset;
2929 #if defined(ia64_HOST_ARCH)
2931 findElfSegment ( void* objImage, Elf_Addr vaddr )
2933 char* ehdrC = (char*)objImage;
2934 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2935 Elf_Phdr* phdr = (Elf_Phdr*)(ehdrC + ehdr->e_phoff);
2936 Elf_Addr segaddr = 0;
2939 for (i = 0; i < ehdr->e_phnum; i++) {
2940 segaddr = phdr[i].p_vaddr;
2941 if ((vaddr >= segaddr) && (vaddr < segaddr + phdr[i].p_memsz))
2949 ocVerifyImage_ELF ( ObjectCode* oc )
2953 int i, j, nent, nstrtab, nsymtabs;
2957 char* ehdrC = (char*)(oc->image);
2958 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2960 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
2961 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
2962 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
2963 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
2964 errorBelch("%s: not an ELF object", oc->fileName);
2968 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
2969 errorBelch("%s: unsupported ELF format", oc->fileName);
2973 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
2974 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
2976 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
2977 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
2979 errorBelch("%s: unknown endiannness", oc->fileName);
2983 if (ehdr->e_type != ET_REL) {
2984 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
2987 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
2989 IF_DEBUG(linker,debugBelch( "Architecture is " ));
2990 switch (ehdr->e_machine) {
2991 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
2992 #ifdef EM_SPARC32PLUS
2993 case EM_SPARC32PLUS:
2995 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
2997 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
2999 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
3001 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
3002 #elif defined(EM_AMD64)
3003 case EM_AMD64: IF_DEBUG(linker,debugBelch( "amd64" )); break;
3005 default: IF_DEBUG(linker,debugBelch( "unknown" ));
3006 errorBelch("%s: unknown architecture (e_machine == %d)"
3007 , oc->fileName, ehdr->e_machine);
3011 IF_DEBUG(linker,debugBelch(
3012 "\nSection header table: start %ld, n_entries %d, ent_size %d\n",
3013 (long)ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
3015 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
3017 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3019 if (ehdr->e_shstrndx == SHN_UNDEF) {
3020 errorBelch("%s: no section header string table", oc->fileName);
3023 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
3025 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3028 for (i = 0; i < ehdr->e_shnum; i++) {
3029 IF_DEBUG(linker,debugBelch("%2d: ", i ));
3030 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
3031 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
3032 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
3033 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
3034 ehdrC + shdr[i].sh_offset,
3035 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
3037 if (shdr[i].sh_type == SHT_REL) {
3038 IF_DEBUG(linker,debugBelch("Rel " ));
3039 } else if (shdr[i].sh_type == SHT_RELA) {
3040 IF_DEBUG(linker,debugBelch("RelA " ));
3042 IF_DEBUG(linker,debugBelch(" "));
3045 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
3049 IF_DEBUG(linker,debugBelch( "\nString tables" ));
3052 for (i = 0; i < ehdr->e_shnum; i++) {
3053 if (shdr[i].sh_type == SHT_STRTAB
3054 /* Ignore the section header's string table. */
3055 && i != ehdr->e_shstrndx
3056 /* Ignore string tables named .stabstr, as they contain
3058 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3060 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
3061 strtab = ehdrC + shdr[i].sh_offset;
3066 errorBelch("%s: no string tables, or too many", oc->fileName);
3071 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
3072 for (i = 0; i < ehdr->e_shnum; i++) {
3073 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3074 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
3076 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3077 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3078 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%ld rem)\n",
3080 (long)shdr[i].sh_size % sizeof(Elf_Sym)
3082 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
3083 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
3086 for (j = 0; j < nent; j++) {
3087 IF_DEBUG(linker,debugBelch(" %2d ", j ));
3088 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
3089 (int)stab[j].st_shndx,
3090 (int)stab[j].st_size,
3091 (char*)stab[j].st_value ));
3093 IF_DEBUG(linker,debugBelch("type=" ));
3094 switch (ELF_ST_TYPE(stab[j].st_info)) {
3095 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
3096 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
3097 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
3098 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
3099 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
3100 default: IF_DEBUG(linker,debugBelch("? " )); break;
3102 IF_DEBUG(linker,debugBelch(" " ));
3104 IF_DEBUG(linker,debugBelch("bind=" ));
3105 switch (ELF_ST_BIND(stab[j].st_info)) {
3106 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
3107 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
3108 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
3109 default: IF_DEBUG(linker,debugBelch("? " )); break;
3111 IF_DEBUG(linker,debugBelch(" " ));
3113 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
3117 if (nsymtabs == 0) {
3118 errorBelch("%s: didn't find any symbol tables", oc->fileName);
3125 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
3129 if (hdr->sh_type == SHT_PROGBITS
3130 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
3131 /* .text-style section */
3132 return SECTIONKIND_CODE_OR_RODATA;
3135 if (hdr->sh_type == SHT_PROGBITS
3136 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3137 /* .data-style section */
3138 return SECTIONKIND_RWDATA;
3141 if (hdr->sh_type == SHT_PROGBITS
3142 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
3143 /* .rodata-style section */
3144 return SECTIONKIND_CODE_OR_RODATA;
3147 if (hdr->sh_type == SHT_NOBITS
3148 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3149 /* .bss-style section */
3151 return SECTIONKIND_RWDATA;
3154 return SECTIONKIND_OTHER;
3159 ocGetNames_ELF ( ObjectCode* oc )
3164 char* ehdrC = (char*)(oc->image);
3165 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3166 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
3167 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3169 ASSERT(symhash != NULL);
3172 errorBelch("%s: no strtab", oc->fileName);
3177 for (i = 0; i < ehdr->e_shnum; i++) {
3178 /* Figure out what kind of section it is. Logic derived from
3179 Figure 1.14 ("Special Sections") of the ELF document
3180 ("Portable Formats Specification, Version 1.1"). */
3182 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
3184 if (is_bss && shdr[i].sh_size > 0) {
3185 /* This is a non-empty .bss section. Allocate zeroed space for
3186 it, and set its .sh_offset field such that
3187 ehdrC + .sh_offset == addr_of_zeroed_space. */
3188 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
3189 "ocGetNames_ELF(BSS)");
3190 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
3192 debugBelch("BSS section at 0x%x, size %d\n",
3193 zspace, shdr[i].sh_size);
3197 /* fill in the section info */
3198 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
3199 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
3200 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
3201 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
3204 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3206 /* copy stuff into this module's object symbol table */
3207 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3208 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3210 oc->n_symbols = nent;
3211 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3212 "ocGetNames_ELF(oc->symbols)");
3214 for (j = 0; j < nent; j++) {
3216 char isLocal = FALSE; /* avoids uninit-var warning */
3218 char* nm = strtab + stab[j].st_name;
3219 int secno = stab[j].st_shndx;
3221 /* Figure out if we want to add it; if so, set ad to its
3222 address. Otherwise leave ad == NULL. */
3224 if (secno == SHN_COMMON) {
3226 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
3228 debugBelch("COMMON symbol, size %d name %s\n",
3229 stab[j].st_size, nm);
3231 /* Pointless to do addProddableBlock() for this area,
3232 since the linker should never poke around in it. */
3235 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
3236 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
3238 /* and not an undefined symbol */
3239 && stab[j].st_shndx != SHN_UNDEF
3240 /* and not in a "special section" */
3241 && stab[j].st_shndx < SHN_LORESERVE
3243 /* and it's a not a section or string table or anything silly */
3244 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
3245 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
3246 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
3249 /* Section 0 is the undefined section, hence > and not >=. */
3250 ASSERT(secno > 0 && secno < ehdr->e_shnum);
3252 if (shdr[secno].sh_type == SHT_NOBITS) {
3253 debugBelch(" BSS symbol, size %d off %d name %s\n",
3254 stab[j].st_size, stab[j].st_value, nm);
3257 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
3258 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
3261 #ifdef ELF_FUNCTION_DESC
3262 /* dlsym() and the initialisation table both give us function
3263 * descriptors, so to be consistent we store function descriptors
3264 * in the symbol table */
3265 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
3266 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
3268 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s\n",
3269 ad, oc->fileName, nm ));
3274 /* And the decision is ... */
3278 oc->symbols[j] = nm;
3281 /* Ignore entirely. */
3283 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
3287 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
3288 strtab + stab[j].st_name ));
3291 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
3292 (int)ELF_ST_BIND(stab[j].st_info),
3293 (int)ELF_ST_TYPE(stab[j].st_info),
3294 (int)stab[j].st_shndx,
3295 strtab + stab[j].st_name
3298 oc->symbols[j] = NULL;
3307 /* Do ELF relocations which lack an explicit addend. All x86-linux
3308 relocations appear to be of this form. */
3310 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
3311 Elf_Shdr* shdr, int shnum,
3312 Elf_Sym* stab, char* strtab )
3317 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
3318 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
3319 int target_shndx = shdr[shnum].sh_info;
3320 int symtab_shndx = shdr[shnum].sh_link;
3322 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3323 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
3324 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3325 target_shndx, symtab_shndx ));
3327 /* Skip sections that we're not interested in. */
3330 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
3331 if (kind == SECTIONKIND_OTHER) {
3332 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
3337 for (j = 0; j < nent; j++) {
3338 Elf_Addr offset = rtab[j].r_offset;
3339 Elf_Addr info = rtab[j].r_info;
3341 Elf_Addr P = ((Elf_Addr)targ) + offset;
3342 Elf_Word* pP = (Elf_Word*)P;
3347 StgStablePtr stablePtr;
3350 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
3351 j, (void*)offset, (void*)info ));
3353 IF_DEBUG(linker,debugBelch( " ZERO" ));
3356 Elf_Sym sym = stab[ELF_R_SYM(info)];
3357 /* First see if it is a local symbol. */
3358 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3359 /* Yes, so we can get the address directly from the ELF symbol
3361 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3363 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3364 + stab[ELF_R_SYM(info)].st_value);
3367 symbol = strtab + sym.st_name;
3368 stablePtr = (StgStablePtr)lookupHashTable(stablehash, (StgWord)symbol);
3369 if (NULL == stablePtr) {
3370 /* No, so look up the name in our global table. */
3371 S_tmp = lookupSymbol( symbol );
3372 S = (Elf_Addr)S_tmp;
3374 stableVal = deRefStablePtr( stablePtr );
3376 S = (Elf_Addr)S_tmp;
3380 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3383 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
3386 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
3387 (void*)P, (void*)S, (void*)A ));
3388 checkProddableBlock ( oc, pP );
3392 switch (ELF_R_TYPE(info)) {
3393 # ifdef i386_HOST_ARCH
3394 case R_386_32: *pP = value; break;
3395 case R_386_PC32: *pP = value - P; break;
3398 errorBelch("%s: unhandled ELF relocation(Rel) type %lu\n",
3399 oc->fileName, (lnat)ELF_R_TYPE(info));
3407 /* Do ELF relocations for which explicit addends are supplied.
3408 sparc-solaris relocations appear to be of this form. */
3410 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
3411 Elf_Shdr* shdr, int shnum,
3412 Elf_Sym* stab, char* strtab )
3415 char *symbol = NULL;
3417 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
3418 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
3419 int target_shndx = shdr[shnum].sh_info;
3420 int symtab_shndx = shdr[shnum].sh_link;
3422 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3423 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
3424 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3425 target_shndx, symtab_shndx ));
3427 for (j = 0; j < nent; j++) {
3428 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3429 /* This #ifdef only serves to avoid unused-var warnings. */
3430 Elf_Addr offset = rtab[j].r_offset;
3431 Elf_Addr P = targ + offset;
3433 Elf_Addr info = rtab[j].r_info;
3434 Elf_Addr A = rtab[j].r_addend;
3438 # if defined(sparc_HOST_ARCH)
3439 Elf_Word* pP = (Elf_Word*)P;
3441 # elif defined(ia64_HOST_ARCH)
3442 Elf64_Xword *pP = (Elf64_Xword *)P;
3444 # elif defined(powerpc_HOST_ARCH)
3448 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
3449 j, (void*)offset, (void*)info,
3452 IF_DEBUG(linker,debugBelch( " ZERO" ));
3455 Elf_Sym sym = stab[ELF_R_SYM(info)];
3456 /* First see if it is a local symbol. */
3457 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3458 /* Yes, so we can get the address directly from the ELF symbol
3460 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3462 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3463 + stab[ELF_R_SYM(info)].st_value);
3464 #ifdef ELF_FUNCTION_DESC
3465 /* Make a function descriptor for this function */
3466 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
3467 S = allocateFunctionDesc(S + A);
3472 /* No, so look up the name in our global table. */
3473 symbol = strtab + sym.st_name;
3474 S_tmp = lookupSymbol( symbol );
3475 S = (Elf_Addr)S_tmp;
3477 #ifdef ELF_FUNCTION_DESC
3478 /* If a function, already a function descriptor - we would
3479 have to copy it to add an offset. */
3480 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3481 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3485 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3488 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3491 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3492 (void*)P, (void*)S, (void*)A ));
3493 /* checkProddableBlock ( oc, (void*)P ); */
3497 switch (ELF_R_TYPE(info)) {
3498 # if defined(sparc_HOST_ARCH)
3499 case R_SPARC_WDISP30:
3500 w1 = *pP & 0xC0000000;
3501 w2 = (Elf_Word)((value - P) >> 2);
3502 ASSERT((w2 & 0xC0000000) == 0);
3507 w1 = *pP & 0xFFC00000;
3508 w2 = (Elf_Word)(value >> 10);
3509 ASSERT((w2 & 0xFFC00000) == 0);
3515 w2 = (Elf_Word)(value & 0x3FF);
3516 ASSERT((w2 & ~0x3FF) == 0);
3520 /* According to the Sun documentation:
3522 This relocation type resembles R_SPARC_32, except it refers to an
3523 unaligned word. That is, the word to be relocated must be treated
3524 as four separate bytes with arbitrary alignment, not as a word
3525 aligned according to the architecture requirements.
3527 (JRS: which means that freeloading on the R_SPARC_32 case
3528 is probably wrong, but hey ...)
3532 w2 = (Elf_Word)value;
3535 # elif defined(ia64_HOST_ARCH)
3536 case R_IA64_DIR64LSB:
3537 case R_IA64_FPTR64LSB:
3540 case R_IA64_PCREL64LSB:
3543 case R_IA64_SEGREL64LSB:
3544 addr = findElfSegment(ehdrC, value);
3547 case R_IA64_GPREL22:
3548 ia64_reloc_gprel22(P, value);
3550 case R_IA64_LTOFF22:
3551 case R_IA64_LTOFF22X:
3552 case R_IA64_LTOFF_FPTR22:
3553 addr = allocateGOTEntry(value);
3554 ia64_reloc_gprel22(P, addr);
3556 case R_IA64_PCREL21B:
3557 ia64_reloc_pcrel21(P, S, oc);
3560 /* This goes with R_IA64_LTOFF22X and points to the load to
3561 * convert into a move. We don't implement relaxation. */
3563 # elif defined(powerpc_HOST_ARCH)
3564 case R_PPC_ADDR16_LO:
3565 *(Elf32_Half*) P = value;
3568 case R_PPC_ADDR16_HI:
3569 *(Elf32_Half*) P = value >> 16;
3572 case R_PPC_ADDR16_HA:
3573 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3577 *(Elf32_Word *) P = value;
3581 *(Elf32_Word *) P = value - P;
3587 if( delta << 6 >> 6 != delta )
3589 value = (Elf_Addr) (&makeSymbolExtra( oc, ELF_R_SYM(info), value )
3593 if( value == 0 || delta << 6 >> 6 != delta )
3595 barf( "Unable to make SymbolExtra for #%d",
3601 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3602 | (delta & 0x3fffffc);
3606 #if x86_64_HOST_ARCH
3608 *(Elf64_Xword *)P = value;
3613 StgInt64 off = value - P;
3614 if (off >= 0x7fffffffL || off < -0x80000000L) {
3615 #if X86_64_ELF_NONPIC_HACK
3616 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3618 off = pltAddress + A - P;
3620 barf("R_X86_64_PC32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3621 symbol, off, oc->fileName );
3624 *(Elf64_Word *)P = (Elf64_Word)off;
3630 StgInt64 off = value - P;
3631 *(Elf64_Word *)P = (Elf64_Word)off;
3636 if (value >= 0x7fffffffL) {
3637 #if X86_64_ELF_NONPIC_HACK
3638 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3640 value = pltAddress + A;
3642 barf("R_X86_64_32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3643 symbol, value, oc->fileName );
3646 *(Elf64_Word *)P = (Elf64_Word)value;
3650 if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) {
3651 #if X86_64_ELF_NONPIC_HACK
3652 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3654 value = pltAddress + A;
3656 barf("R_X86_64_32S relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3657 symbol, value, oc->fileName );
3660 *(Elf64_Sword *)P = (Elf64_Sword)value;
3663 case R_X86_64_GOTPCREL:
3665 StgInt64 gotAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)->addr;
3666 StgInt64 off = gotAddress + A - P;
3667 *(Elf64_Word *)P = (Elf64_Word)off;
3671 case R_X86_64_PLT32:
3673 StgInt64 off = value - P;
3674 if (off >= 0x7fffffffL || off < -0x80000000L) {
3675 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3677 off = pltAddress + A - P;
3679 *(Elf64_Word *)P = (Elf64_Word)off;
3685 errorBelch("%s: unhandled ELF relocation(RelA) type %lu\n",
3686 oc->fileName, (lnat)ELF_R_TYPE(info));
3695 ocResolve_ELF ( ObjectCode* oc )
3699 Elf_Sym* stab = NULL;
3700 char* ehdrC = (char*)(oc->image);
3701 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3702 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3704 /* first find "the" symbol table */
3705 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3707 /* also go find the string table */
3708 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3710 if (stab == NULL || strtab == NULL) {
3711 errorBelch("%s: can't find string or symbol table", oc->fileName);
3715 /* Process the relocation sections. */
3716 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3717 if (shdr[shnum].sh_type == SHT_REL) {
3718 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3719 shnum, stab, strtab );
3723 if (shdr[shnum].sh_type == SHT_RELA) {
3724 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3725 shnum, stab, strtab );
3730 #if defined(powerpc_HOST_ARCH)
3731 ocFlushInstructionCache( oc );
3739 * Instructions are 41 bits long, packed into 128 bit bundles with a 5-bit template
3740 * at the front. The following utility functions pack and unpack instructions, and
3741 * take care of the most common relocations.
3744 #ifdef ia64_HOST_ARCH
3747 ia64_extract_instruction(Elf64_Xword *target)
3750 int slot = (Elf_Addr)target & 3;
3751 target = (Elf_Addr)target & ~3;
3759 return ((w1 >> 5) & 0x1ffffffffff);
3761 return (w1 >> 46) | ((w2 & 0x7fffff) << 18);
3765 barf("ia64_extract_instruction: invalid slot %p", target);
3770 ia64_deposit_instruction(Elf64_Xword *target, Elf64_Xword value)
3772 int slot = (Elf_Addr)target & 3;
3773 target = (Elf_Addr)target & ~3;
3778 *target |= value << 5;
3781 *target |= value << 46;
3782 *(target+1) |= value >> 18;
3785 *(target+1) |= value << 23;
3791 ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value)
3793 Elf64_Xword instruction;
3794 Elf64_Sxword rel_value;
3796 rel_value = value - gp_val;
3797 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff))
3798 barf("GP-relative data out of range (address = 0x%lx, gp = 0x%lx)", value, gp_val);
3800 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3801 instruction |= (((rel_value >> 0) & 0x07f) << 13) /* imm7b */
3802 | (((rel_value >> 7) & 0x1ff) << 27) /* imm9d */
3803 | (((rel_value >> 16) & 0x01f) << 22) /* imm5c */
3804 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3805 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3809 ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc)
3811 Elf64_Xword instruction;
3812 Elf64_Sxword rel_value;
3815 entry = allocatePLTEntry(value, oc);
3817 rel_value = (entry >> 4) - (target >> 4);
3818 if ((rel_value > 0xfffff) || (rel_value < -0xfffff))
3819 barf("PLT entry too far away (entry = 0x%lx, target = 0x%lx)", entry, target);
3821 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3822 instruction |= ((rel_value & 0xfffff) << 13) /* imm20b */
3823 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3824 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3830 * PowerPC & X86_64 ELF specifics
3833 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3835 static int ocAllocateSymbolExtras_ELF( ObjectCode *oc )
3841 ehdr = (Elf_Ehdr *) oc->image;
3842 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3844 for( i = 0; i < ehdr->e_shnum; i++ )
3845 if( shdr[i].sh_type == SHT_SYMTAB )
3848 if( i == ehdr->e_shnum )
3850 errorBelch( "This ELF file contains no symtab" );
3854 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3856 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3857 (int) shdr[i].sh_entsize, (int) sizeof( Elf_Sym ) );
3862 return ocAllocateSymbolExtras( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3865 #endif /* powerpc */
3869 /* --------------------------------------------------------------------------
3871 * ------------------------------------------------------------------------*/
3873 #if defined(OBJFORMAT_MACHO)
3876 Support for MachO linking on Darwin/MacOS X
3877 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3879 I hereby formally apologize for the hackish nature of this code.
3880 Things that need to be done:
3881 *) implement ocVerifyImage_MachO
3882 *) add still more sanity checks.
3885 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
3886 #define mach_header mach_header_64
3887 #define segment_command segment_command_64
3888 #define section section_64
3889 #define nlist nlist_64
3892 #ifdef powerpc_HOST_ARCH
3893 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3895 struct mach_header *header = (struct mach_header *) oc->image;
3896 struct load_command *lc = (struct load_command *) (header + 1);
3899 for( i = 0; i < header->ncmds; i++ )
3901 if( lc->cmd == LC_SYMTAB )
3903 // Find out the first and last undefined external
3904 // symbol, so we don't have to allocate too many
3906 struct symtab_command *symLC = (struct symtab_command *) lc;
3907 unsigned min = symLC->nsyms, max = 0;
3908 struct nlist *nlist =
3909 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
3911 for(i=0;i<symLC->nsyms;i++)
3913 if(nlist[i].n_type & N_STAB)
3915 else if(nlist[i].n_type & N_EXT)
3917 if((nlist[i].n_type & N_TYPE) == N_UNDF
3918 && (nlist[i].n_value == 0))
3928 return ocAllocateSymbolExtras(oc, max - min + 1, min);
3933 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3935 return ocAllocateSymbolExtras(oc,0,0);
3938 #ifdef x86_64_HOST_ARCH
3939 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3941 struct mach_header *header = (struct mach_header *) oc->image;
3942 struct load_command *lc = (struct load_command *) (header + 1);
3945 for( i = 0; i < header->ncmds; i++ )
3947 if( lc->cmd == LC_SYMTAB )
3949 // Just allocate one entry for every symbol
3950 struct symtab_command *symLC = (struct symtab_command *) lc;
3952 return ocAllocateSymbolExtras(oc, symLC->nsyms, 0);
3955 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3957 return ocAllocateSymbolExtras(oc,0,0);
3961 static int ocVerifyImage_MachO(ObjectCode* oc)
3963 char *image = (char*) oc->image;
3964 struct mach_header *header = (struct mach_header*) image;
3966 #if x86_64_TARGET_ARCH || powerpc64_TARGET_ARCH
3967 if(header->magic != MH_MAGIC_64)
3970 if(header->magic != MH_MAGIC)
3973 // FIXME: do some more verifying here
3977 static int resolveImports(
3980 struct symtab_command *symLC,
3981 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
3982 unsigned long *indirectSyms,
3983 struct nlist *nlist)
3986 size_t itemSize = 4;
3989 int isJumpTable = 0;
3990 if(!strcmp(sect->sectname,"__jump_table"))
3994 ASSERT(sect->reserved2 == itemSize);
3998 for(i=0; i*itemSize < sect->size;i++)
4000 // according to otool, reserved1 contains the first index into the indirect symbol table
4001 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
4002 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4005 if((symbol->n_type & N_TYPE) == N_UNDF
4006 && (symbol->n_type & N_EXT) && (symbol->n_value != 0))
4007 addr = (void*) (symbol->n_value);
4009 addr = lookupSymbol(nm);
4012 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
4020 checkProddableBlock(oc,image + sect->offset + i*itemSize);
4021 *(image + sect->offset + i*itemSize) = 0xe9; // jmp
4022 *(unsigned*)(image + sect->offset + i*itemSize + 1)
4023 = (char*)addr - (image + sect->offset + i*itemSize + 5);
4028 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
4029 ((void**)(image + sect->offset))[i] = addr;
4036 static unsigned long relocateAddress(
4039 struct section* sections,
4040 unsigned long address)
4043 for(i = 0; i < nSections; i++)
4045 if(sections[i].addr <= address
4046 && address < sections[i].addr + sections[i].size)
4048 return (unsigned long)oc->image
4049 + sections[i].offset + address - sections[i].addr;
4052 barf("Invalid Mach-O file:"
4053 "Address out of bounds while relocating object file");
4057 static int relocateSection(
4060 struct symtab_command *symLC, struct nlist *nlist,
4061 int nSections, struct section* sections, struct section *sect)
4063 struct relocation_info *relocs;
4066 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
4068 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
4070 else if(!strcmp(sect->sectname,"__la_sym_ptr2"))
4072 else if(!strcmp(sect->sectname,"__la_sym_ptr3"))
4076 relocs = (struct relocation_info*) (image + sect->reloff);
4080 #ifdef x86_64_HOST_ARCH
4081 struct relocation_info *reloc = &relocs[i];
4083 char *thingPtr = image + sect->offset + reloc->r_address;
4087 int type = reloc->r_type;
4089 checkProddableBlock(oc,thingPtr);
4090 switch(reloc->r_length)
4093 thing = *(uint8_t*)thingPtr;
4094 baseValue = (uint64_t)thingPtr + 1;
4097 thing = *(uint16_t*)thingPtr;
4098 baseValue = (uint64_t)thingPtr + 2;
4101 thing = *(uint32_t*)thingPtr;
4102 baseValue = (uint64_t)thingPtr + 4;
4105 thing = *(uint64_t*)thingPtr;
4106 baseValue = (uint64_t)thingPtr + 8;
4109 barf("Unknown size.");
4112 if(type == X86_64_RELOC_GOT
4113 || type == X86_64_RELOC_GOT_LOAD)
4115 ASSERT(reloc->r_extern);
4116 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)->addr;
4118 type = X86_64_RELOC_SIGNED;
4120 else if(reloc->r_extern)
4122 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4123 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4124 if(symbol->n_value == 0)
4125 value = (uint64_t) lookupSymbol(nm);
4127 value = relocateAddress(oc, nSections, sections,
4132 value = sections[reloc->r_symbolnum-1].offset
4133 - sections[reloc->r_symbolnum-1].addr
4137 if(type == X86_64_RELOC_BRANCH)
4139 if((int32_t)(value - baseValue) != (int64_t)(value - baseValue))
4141 ASSERT(reloc->r_extern);
4142 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)
4145 ASSERT((int32_t)(value - baseValue) == (int64_t)(value - baseValue));
4146 type = X86_64_RELOC_SIGNED;
4151 case X86_64_RELOC_UNSIGNED:
4152 ASSERT(!reloc->r_pcrel);
4155 case X86_64_RELOC_SIGNED:
4156 ASSERT(reloc->r_pcrel);
4157 thing += value - baseValue;
4159 case X86_64_RELOC_SUBTRACTOR:
4160 ASSERT(!reloc->r_pcrel);
4164 barf("unkown relocation");
4167 switch(reloc->r_length)
4170 *(uint8_t*)thingPtr = thing;
4173 *(uint16_t*)thingPtr = thing;
4176 *(uint32_t*)thingPtr = thing;
4179 *(uint64_t*)thingPtr = thing;
4183 if(relocs[i].r_address & R_SCATTERED)
4185 struct scattered_relocation_info *scat =
4186 (struct scattered_relocation_info*) &relocs[i];
4190 if(scat->r_length == 2)
4192 unsigned long word = 0;
4193 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
4194 checkProddableBlock(oc,wordPtr);
4196 // Note on relocation types:
4197 // i386 uses the GENERIC_RELOC_* types,
4198 // while ppc uses special PPC_RELOC_* types.
4199 // *_RELOC_VANILLA and *_RELOC_PAIR have the same value
4200 // in both cases, all others are different.
4201 // Therefore, we use GENERIC_RELOC_VANILLA
4202 // and GENERIC_RELOC_PAIR instead of the PPC variants,
4203 // and use #ifdefs for the other types.
4205 // Step 1: Figure out what the relocated value should be
4206 if(scat->r_type == GENERIC_RELOC_VANILLA)
4208 word = *wordPtr + (unsigned long) relocateAddress(
4215 #ifdef powerpc_HOST_ARCH
4216 else if(scat->r_type == PPC_RELOC_SECTDIFF
4217 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
4218 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
4219 || scat->r_type == PPC_RELOC_HA16_SECTDIFF)
4221 else if(scat->r_type == GENERIC_RELOC_SECTDIFF)
4224 struct scattered_relocation_info *pair =
4225 (struct scattered_relocation_info*) &relocs[i+1];
4227 if(!pair->r_scattered || pair->r_type != GENERIC_RELOC_PAIR)
4228 barf("Invalid Mach-O file: "
4229 "RELOC_*_SECTDIFF not followed by RELOC_PAIR");
4231 word = (unsigned long)
4232 (relocateAddress(oc, nSections, sections, scat->r_value)
4233 - relocateAddress(oc, nSections, sections, pair->r_value));
4236 #ifdef powerpc_HOST_ARCH
4237 else if(scat->r_type == PPC_RELOC_HI16
4238 || scat->r_type == PPC_RELOC_LO16
4239 || scat->r_type == PPC_RELOC_HA16
4240 || scat->r_type == PPC_RELOC_LO14)
4241 { // these are generated by label+offset things
4242 struct relocation_info *pair = &relocs[i+1];
4243 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
4244 barf("Invalid Mach-O file: "
4245 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
4247 if(scat->r_type == PPC_RELOC_LO16)
4249 word = ((unsigned short*) wordPtr)[1];
4250 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4252 else if(scat->r_type == PPC_RELOC_LO14)
4254 barf("Unsupported Relocation: PPC_RELOC_LO14");
4255 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
4256 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4258 else if(scat->r_type == PPC_RELOC_HI16)
4260 word = ((unsigned short*) wordPtr)[1] << 16;
4261 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4263 else if(scat->r_type == PPC_RELOC_HA16)
4265 word = ((unsigned short*) wordPtr)[1] << 16;
4266 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4270 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
4277 continue; // ignore the others
4279 #ifdef powerpc_HOST_ARCH
4280 if(scat->r_type == GENERIC_RELOC_VANILLA
4281 || scat->r_type == PPC_RELOC_SECTDIFF)
4283 if(scat->r_type == GENERIC_RELOC_VANILLA
4284 || scat->r_type == GENERIC_RELOC_SECTDIFF)
4289 #ifdef powerpc_HOST_ARCH
4290 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
4292 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4294 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
4296 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4298 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
4300 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4301 + ((word & (1<<15)) ? 1 : 0);
4307 continue; // FIXME: I hope it's OK to ignore all the others.
4311 struct relocation_info *reloc = &relocs[i];
4312 if(reloc->r_pcrel && !reloc->r_extern)
4315 if(reloc->r_length == 2)
4317 unsigned long word = 0;
4318 #ifdef powerpc_HOST_ARCH
4319 unsigned long jumpIsland = 0;
4320 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
4321 // to avoid warning and to catch
4325 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
4326 checkProddableBlock(oc,wordPtr);
4328 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4332 #ifdef powerpc_HOST_ARCH
4333 else if(reloc->r_type == PPC_RELOC_LO16)
4335 word = ((unsigned short*) wordPtr)[1];
4336 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4338 else if(reloc->r_type == PPC_RELOC_HI16)
4340 word = ((unsigned short*) wordPtr)[1] << 16;
4341 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4343 else if(reloc->r_type == PPC_RELOC_HA16)
4345 word = ((unsigned short*) wordPtr)[1] << 16;
4346 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4348 else if(reloc->r_type == PPC_RELOC_BR24)
4351 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
4355 if(!reloc->r_extern)
4358 sections[reloc->r_symbolnum-1].offset
4359 - sections[reloc->r_symbolnum-1].addr
4366 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4367 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4368 void *symbolAddress = lookupSymbol(nm);
4371 errorBelch("\nunknown symbol `%s'", nm);
4377 #ifdef powerpc_HOST_ARCH
4378 // In the .o file, this should be a relative jump to NULL
4379 // and we'll change it to a relative jump to the symbol
4380 ASSERT(word + reloc->r_address == 0);
4381 jumpIsland = (unsigned long)
4382 &makeSymbolExtra(oc,
4384 (unsigned long) symbolAddress)
4388 offsetToJumpIsland = word + jumpIsland
4389 - (((long)image) + sect->offset - sect->addr);
4392 word += (unsigned long) symbolAddress
4393 - (((long)image) + sect->offset - sect->addr);
4397 word += (unsigned long) symbolAddress;
4401 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4406 #ifdef powerpc_HOST_ARCH
4407 else if(reloc->r_type == PPC_RELOC_LO16)
4409 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4412 else if(reloc->r_type == PPC_RELOC_HI16)
4414 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4417 else if(reloc->r_type == PPC_RELOC_HA16)
4419 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4420 + ((word & (1<<15)) ? 1 : 0);
4423 else if(reloc->r_type == PPC_RELOC_BR24)
4425 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4427 // The branch offset is too large.
4428 // Therefore, we try to use a jump island.
4431 barf("unconditional relative branch out of range: "
4432 "no jump island available");
4435 word = offsetToJumpIsland;
4436 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4437 barf("unconditional relative branch out of range: "
4438 "jump island out of range");
4440 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
4445 barf("\nunknown relocation %d",reloc->r_type);
4453 static int ocGetNames_MachO(ObjectCode* oc)
4455 char *image = (char*) oc->image;
4456 struct mach_header *header = (struct mach_header*) image;
4457 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4458 unsigned i,curSymbol = 0;
4459 struct segment_command *segLC = NULL;
4460 struct section *sections;
4461 struct symtab_command *symLC = NULL;
4462 struct nlist *nlist;
4463 unsigned long commonSize = 0;
4464 char *commonStorage = NULL;
4465 unsigned long commonCounter;
4467 for(i=0;i<header->ncmds;i++)
4469 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4470 segLC = (struct segment_command*) lc;
4471 else if(lc->cmd == LC_SYMTAB)
4472 symLC = (struct symtab_command*) lc;
4473 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4476 sections = (struct section*) (segLC+1);
4477 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4481 barf("ocGetNames_MachO: no segment load command");
4483 for(i=0;i<segLC->nsects;i++)
4485 if(sections[i].size == 0)
4488 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
4490 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
4491 "ocGetNames_MachO(common symbols)");
4492 sections[i].offset = zeroFillArea - image;
4495 if(!strcmp(sections[i].sectname,"__text"))
4496 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
4497 (void*) (image + sections[i].offset),
4498 (void*) (image + sections[i].offset + sections[i].size));
4499 else if(!strcmp(sections[i].sectname,"__const"))
4500 addSection(oc, SECTIONKIND_RWDATA,
4501 (void*) (image + sections[i].offset),
4502 (void*) (image + sections[i].offset + sections[i].size));
4503 else if(!strcmp(sections[i].sectname,"__data"))
4504 addSection(oc, SECTIONKIND_RWDATA,
4505 (void*) (image + sections[i].offset),
4506 (void*) (image + sections[i].offset + sections[i].size));
4507 else if(!strcmp(sections[i].sectname,"__bss")
4508 || !strcmp(sections[i].sectname,"__common"))
4509 addSection(oc, SECTIONKIND_RWDATA,
4510 (void*) (image + sections[i].offset),
4511 (void*) (image + sections[i].offset + sections[i].size));
4513 addProddableBlock(oc, (void*) (image + sections[i].offset),
4517 // count external symbols defined here
4521 for(i=0;i<symLC->nsyms;i++)
4523 if(nlist[i].n_type & N_STAB)
4525 else if(nlist[i].n_type & N_EXT)
4527 if((nlist[i].n_type & N_TYPE) == N_UNDF
4528 && (nlist[i].n_value != 0))
4530 commonSize += nlist[i].n_value;
4533 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4538 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
4539 "ocGetNames_MachO(oc->symbols)");
4543 for(i=0;i<symLC->nsyms;i++)
4545 if(nlist[i].n_type & N_STAB)
4547 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4549 if(nlist[i].n_type & N_EXT)
4551 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4552 if((nlist[i].n_desc & N_WEAK_DEF) && lookupSymbol(nm))
4553 ; // weak definition, and we already have a definition
4556 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4558 + sections[nlist[i].n_sect-1].offset
4559 - sections[nlist[i].n_sect-1].addr
4560 + nlist[i].n_value);
4561 oc->symbols[curSymbol++] = nm;
4568 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
4569 commonCounter = (unsigned long)commonStorage;
4572 for(i=0;i<symLC->nsyms;i++)
4574 if((nlist[i].n_type & N_TYPE) == N_UNDF
4575 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
4577 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4578 unsigned long sz = nlist[i].n_value;
4580 nlist[i].n_value = commonCounter;
4582 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4583 (void*)commonCounter);
4584 oc->symbols[curSymbol++] = nm;
4586 commonCounter += sz;
4593 static int ocResolve_MachO(ObjectCode* oc)
4595 char *image = (char*) oc->image;
4596 struct mach_header *header = (struct mach_header*) image;
4597 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4599 struct segment_command *segLC = NULL;
4600 struct section *sections;
4601 struct symtab_command *symLC = NULL;
4602 struct dysymtab_command *dsymLC = NULL;
4603 struct nlist *nlist;
4605 for(i=0;i<header->ncmds;i++)
4607 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4608 segLC = (struct segment_command*) lc;
4609 else if(lc->cmd == LC_SYMTAB)
4610 symLC = (struct symtab_command*) lc;
4611 else if(lc->cmd == LC_DYSYMTAB)
4612 dsymLC = (struct dysymtab_command*) lc;
4613 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4616 sections = (struct section*) (segLC+1);
4617 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4622 unsigned long *indirectSyms
4623 = (unsigned long*) (image + dsymLC->indirectsymoff);
4625 for(i=0;i<segLC->nsects;i++)
4627 if( !strcmp(sections[i].sectname,"__la_symbol_ptr")
4628 || !strcmp(sections[i].sectname,"__la_sym_ptr2")
4629 || !strcmp(sections[i].sectname,"__la_sym_ptr3"))
4631 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4634 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr")
4635 || !strcmp(sections[i].sectname,"__pointers"))
4637 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4640 else if(!strcmp(sections[i].sectname,"__jump_table"))
4642 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4648 for(i=0;i<segLC->nsects;i++)
4650 if(!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
4654 #if defined (powerpc_HOST_ARCH)
4655 ocFlushInstructionCache( oc );
4661 #ifdef powerpc_HOST_ARCH
4663 * The Mach-O object format uses leading underscores. But not everywhere.
4664 * There is a small number of runtime support functions defined in
4665 * libcc_dynamic.a whose name does not have a leading underscore.
4666 * As a consequence, we can't get their address from C code.
4667 * We have to use inline assembler just to take the address of a function.
4671 static void machoInitSymbolsWithoutUnderscore()
4673 extern void* symbolsWithoutUnderscore[];
4674 void **p = symbolsWithoutUnderscore;
4675 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
4677 #undef SymI_NeedsProto
4678 #define SymI_NeedsProto(x) \
4679 __asm__ volatile(".long " # x);
4681 RTS_MACHO_NOUNDERLINE_SYMBOLS
4683 __asm__ volatile(".text");
4685 #undef SymI_NeedsProto
4686 #define SymI_NeedsProto(x) \
4687 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
4689 RTS_MACHO_NOUNDERLINE_SYMBOLS
4691 #undef SymI_NeedsProto
4696 * Figure out by how much to shift the entire Mach-O file in memory
4697 * when loading so that its single segment ends up 16-byte-aligned
4699 static int machoGetMisalignment( FILE * f )
4701 struct mach_header header;
4704 fread(&header, sizeof(header), 1, f);
4707 #if x86_64_TARGET_ARCH || powerpc64_TARGET_ARCH
4708 if(header.magic != MH_MAGIC_64)
4711 if(header.magic != MH_MAGIC)
4715 misalignment = (header.sizeofcmds + sizeof(header))
4718 return misalignment ? (16 - misalignment) : 0;