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 */
1468 /* load the image into memory */
1469 f = fopen(path, "rb");
1471 barf("loadObj: can't read `%s'", path);
1473 # if defined(mingw32_HOST_OS)
1474 // TODO: We would like to use allocateExec here, but allocateExec
1475 // cannot currently allocate blocks large enough.
1476 oc->image = VirtualAlloc(NULL, oc->fileSize, MEM_RESERVE | MEM_COMMIT,
1477 PAGE_EXECUTE_READWRITE);
1478 # elif defined(darwin_HOST_OS)
1479 // In a Mach-O .o file, all sections can and will be misaligned
1480 // if the total size of the headers is not a multiple of the
1481 // desired alignment. This is fine for .o files that only serve
1482 // as input for the static linker, but it's not fine for us,
1483 // as SSE (used by gcc for floating point) and Altivec require
1484 // 16-byte alignment.
1485 // We calculate the correct alignment from the header before
1486 // reading the file, and then we misalign oc->image on purpose so
1487 // that the actual sections end up aligned again.
1488 oc->misalignment = machoGetMisalignment(f);
1489 oc->image = stgMallocBytes(oc->fileSize + oc->misalignment, "loadObj(image)");
1490 oc->image += oc->misalignment;
1492 oc->image = stgMallocBytes(oc->fileSize, "loadObj(image)");
1495 n = fread ( oc->image, 1, oc->fileSize, f );
1496 if (n != oc->fileSize)
1497 barf("loadObj: error whilst reading `%s'", path);
1500 #endif /* USE_MMAP */
1502 # if defined(OBJFORMAT_MACHO) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1503 r = ocAllocateSymbolExtras_MachO ( oc );
1504 if (!r) { return r; }
1505 # elif defined(OBJFORMAT_ELF) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1506 r = ocAllocateSymbolExtras_ELF ( oc );
1507 if (!r) { return r; }
1510 /* verify the in-memory image */
1511 # if defined(OBJFORMAT_ELF)
1512 r = ocVerifyImage_ELF ( oc );
1513 # elif defined(OBJFORMAT_PEi386)
1514 r = ocVerifyImage_PEi386 ( oc );
1515 # elif defined(OBJFORMAT_MACHO)
1516 r = ocVerifyImage_MachO ( oc );
1518 barf("loadObj: no verify method");
1520 if (!r) { return r; }
1522 /* build the symbol list for this image */
1523 # if defined(OBJFORMAT_ELF)
1524 r = ocGetNames_ELF ( oc );
1525 # elif defined(OBJFORMAT_PEi386)
1526 r = ocGetNames_PEi386 ( oc );
1527 # elif defined(OBJFORMAT_MACHO)
1528 r = ocGetNames_MachO ( oc );
1530 barf("loadObj: no getNames method");
1532 if (!r) { return r; }
1534 /* loaded, but not resolved yet */
1535 oc->status = OBJECT_LOADED;
1540 /* -----------------------------------------------------------------------------
1541 * resolve all the currently unlinked objects in memory
1543 * Returns: 1 if ok, 0 on error.
1553 for (oc = objects; oc; oc = oc->next) {
1554 if (oc->status != OBJECT_RESOLVED) {
1555 # if defined(OBJFORMAT_ELF)
1556 r = ocResolve_ELF ( oc );
1557 # elif defined(OBJFORMAT_PEi386)
1558 r = ocResolve_PEi386 ( oc );
1559 # elif defined(OBJFORMAT_MACHO)
1560 r = ocResolve_MachO ( oc );
1562 barf("resolveObjs: not implemented on this platform");
1564 if (!r) { return r; }
1565 oc->status = OBJECT_RESOLVED;
1571 /* -----------------------------------------------------------------------------
1572 * delete an object from the pool
1575 unloadObj( char *path )
1577 ObjectCode *oc, *prev;
1579 ASSERT(symhash != NULL);
1580 ASSERT(objects != NULL);
1585 for (oc = objects; oc; prev = oc, oc = oc->next) {
1586 if (!strcmp(oc->fileName,path)) {
1588 /* Remove all the mappings for the symbols within this
1593 for (i = 0; i < oc->n_symbols; i++) {
1594 if (oc->symbols[i] != NULL) {
1595 removeStrHashTable(symhash, oc->symbols[i], NULL);
1603 prev->next = oc->next;
1606 // We're going to leave this in place, in case there are
1607 // any pointers from the heap into it:
1608 // #ifdef mingw32_HOST_OS
1609 // VirtualFree(oc->image);
1611 // stgFree(oc->image);
1613 stgFree(oc->fileName);
1614 stgFree(oc->symbols);
1615 stgFree(oc->sections);
1621 errorBelch("unloadObj: can't find `%s' to unload", path);
1625 /* -----------------------------------------------------------------------------
1626 * Sanity checking. For each ObjectCode, maintain a list of address ranges
1627 * which may be prodded during relocation, and abort if we try and write
1628 * outside any of these.
1630 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
1633 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
1634 /* debugBelch("aPB %p %p %d\n", oc, start, size); */
1638 pb->next = oc->proddables;
1639 oc->proddables = pb;
1642 static void checkProddableBlock ( ObjectCode* oc, void* addr )
1645 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
1646 char* s = (char*)(pb->start);
1647 char* e = s + pb->size - 1;
1648 char* a = (char*)addr;
1649 /* Assumes that the biggest fixup involves a 4-byte write. This
1650 probably needs to be changed to 8 (ie, +7) on 64-bit
1652 if (a >= s && (a+3) <= e) return;
1654 barf("checkProddableBlock: invalid fixup in runtime linker");
1657 /* -----------------------------------------------------------------------------
1658 * Section management.
1660 static void addSection ( ObjectCode* oc, SectionKind kind,
1661 void* start, void* end )
1663 Section* s = stgMallocBytes(sizeof(Section), "addSection");
1667 s->next = oc->sections;
1670 debugBelch("addSection: %p-%p (size %d), kind %d\n",
1671 start, ((char*)end)-1, end - start + 1, kind );
1676 /* --------------------------------------------------------------------------
1678 * This is about allocating a small chunk of memory for every symbol in the
1679 * object file. We make sure that the SymboLExtras are always "in range" of
1680 * limited-range PC-relative instructions on various platforms by allocating
1681 * them right next to the object code itself.
1684 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
1687 ocAllocateSymbolExtras
1689 Allocate additional space at the end of the object file image to make room
1690 for jump islands (powerpc, x86_64) and GOT entries (x86_64).
1692 PowerPC relative branch instructions have a 24 bit displacement field.
1693 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
1694 If a particular imported symbol is outside this range, we have to redirect
1695 the jump to a short piece of new code that just loads the 32bit absolute
1696 address and jumps there.
1697 On x86_64, PC-relative jumps and PC-relative accesses to the GOT are limited
1700 This function just allocates space for one SymbolExtra for every
1701 undefined symbol in the object file. The code for the jump islands is
1702 filled in by makeSymbolExtra below.
1705 static int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first )
1712 int misalignment = 0;
1713 #ifdef darwin_HOST_OS
1714 misalignment = oc->misalignment;
1720 // round up to the nearest 4
1721 aligned = (oc->fileSize + 3) & ~3;
1724 pagesize = getpagesize();
1725 n = ROUND_UP( oc->fileSize, pagesize );
1726 m = ROUND_UP( aligned + sizeof (SymbolExtra) * count, pagesize );
1728 /* we try to use spare space at the end of the last page of the
1729 * image for the jump islands, but if there isn't enough space
1730 * then we have to map some (anonymously, remembering MAP_32BIT).
1732 if( m > n ) // we need to allocate more pages
1734 oc->symbol_extras = mmapForLinker(sizeof(SymbolExtra) * count,
1739 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1742 oc->image -= misalignment;
1743 oc->image = stgReallocBytes( oc->image,
1745 aligned + sizeof (SymbolExtra) * count,
1746 "ocAllocateSymbolExtras" );
1747 oc->image += misalignment;
1749 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1750 #endif /* USE_MMAP */
1752 memset( oc->symbol_extras, 0, sizeof (SymbolExtra) * count );
1755 oc->symbol_extras = NULL;
1757 oc->first_symbol_extra = first;
1758 oc->n_symbol_extras = count;
1763 static SymbolExtra* makeSymbolExtra( ObjectCode* oc,
1764 unsigned long symbolNumber,
1765 unsigned long target )
1769 ASSERT( symbolNumber >= oc->first_symbol_extra
1770 && symbolNumber - oc->first_symbol_extra < oc->n_symbol_extras);
1772 extra = &oc->symbol_extras[symbolNumber - oc->first_symbol_extra];
1774 #ifdef powerpc_HOST_ARCH
1775 // lis r12, hi16(target)
1776 extra->jumpIsland.lis_r12 = 0x3d80;
1777 extra->jumpIsland.hi_addr = target >> 16;
1779 // ori r12, r12, lo16(target)
1780 extra->jumpIsland.ori_r12_r12 = 0x618c;
1781 extra->jumpIsland.lo_addr = target & 0xffff;
1784 extra->jumpIsland.mtctr_r12 = 0x7d8903a6;
1787 extra->jumpIsland.bctr = 0x4e800420;
1789 #ifdef x86_64_HOST_ARCH
1791 static uint8_t jmp[] = { 0xFF, 0x25, 0xF2, 0xFF, 0xFF, 0xFF };
1792 extra->addr = target;
1793 memcpy(extra->jumpIsland, jmp, 6);
1801 /* --------------------------------------------------------------------------
1802 * PowerPC specifics (instruction cache flushing)
1803 * ------------------------------------------------------------------------*/
1805 #ifdef powerpc_TARGET_ARCH
1807 ocFlushInstructionCache
1809 Flush the data & instruction caches.
1810 Because the PPC has split data/instruction caches, we have to
1811 do that whenever we modify code at runtime.
1814 static void ocFlushInstructionCache( ObjectCode *oc )
1816 int n = (oc->fileSize + sizeof( SymbolExtra ) * oc->n_symbol_extras + 3) / 4;
1817 unsigned long *p = (unsigned long *) oc->image;
1821 __asm__ volatile ( "dcbf 0,%0\n\t"
1829 __asm__ volatile ( "sync\n\t"
1835 /* --------------------------------------------------------------------------
1836 * PEi386 specifics (Win32 targets)
1837 * ------------------------------------------------------------------------*/
1839 /* The information for this linker comes from
1840 Microsoft Portable Executable
1841 and Common Object File Format Specification
1842 revision 5.1 January 1998
1843 which SimonM says comes from the MS Developer Network CDs.
1845 It can be found there (on older CDs), but can also be found
1848 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
1850 (this is Rev 6.0 from February 1999).
1852 Things move, so if that fails, try searching for it via
1854 http://www.google.com/search?q=PE+COFF+specification
1856 The ultimate reference for the PE format is the Winnt.h
1857 header file that comes with the Platform SDKs; as always,
1858 implementations will drift wrt their documentation.
1860 A good background article on the PE format is Matt Pietrek's
1861 March 1994 article in Microsoft System Journal (MSJ)
1862 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
1863 Win32 Portable Executable File Format." The info in there
1864 has recently been updated in a two part article in
1865 MSDN magazine, issues Feb and March 2002,
1866 "Inside Windows: An In-Depth Look into the Win32 Portable
1867 Executable File Format"
1869 John Levine's book "Linkers and Loaders" contains useful
1874 #if defined(OBJFORMAT_PEi386)
1878 typedef unsigned char UChar;
1879 typedef unsigned short UInt16;
1880 typedef unsigned int UInt32;
1887 UInt16 NumberOfSections;
1888 UInt32 TimeDateStamp;
1889 UInt32 PointerToSymbolTable;
1890 UInt32 NumberOfSymbols;
1891 UInt16 SizeOfOptionalHeader;
1892 UInt16 Characteristics;
1896 #define sizeof_COFF_header 20
1903 UInt32 VirtualAddress;
1904 UInt32 SizeOfRawData;
1905 UInt32 PointerToRawData;
1906 UInt32 PointerToRelocations;
1907 UInt32 PointerToLinenumbers;
1908 UInt16 NumberOfRelocations;
1909 UInt16 NumberOfLineNumbers;
1910 UInt32 Characteristics;
1914 #define sizeof_COFF_section 40
1921 UInt16 SectionNumber;
1924 UChar NumberOfAuxSymbols;
1928 #define sizeof_COFF_symbol 18
1933 UInt32 VirtualAddress;
1934 UInt32 SymbolTableIndex;
1939 #define sizeof_COFF_reloc 10
1942 /* From PE spec doc, section 3.3.2 */
1943 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
1944 windows.h -- for the same purpose, but I want to know what I'm
1946 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
1947 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
1948 #define MYIMAGE_FILE_DLL 0x2000
1949 #define MYIMAGE_FILE_SYSTEM 0x1000
1950 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
1951 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
1952 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
1954 /* From PE spec doc, section 5.4.2 and 5.4.4 */
1955 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
1956 #define MYIMAGE_SYM_CLASS_STATIC 3
1957 #define MYIMAGE_SYM_UNDEFINED 0
1959 /* From PE spec doc, section 4.1 */
1960 #define MYIMAGE_SCN_CNT_CODE 0x00000020
1961 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
1962 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
1964 /* From PE spec doc, section 5.2.1 */
1965 #define MYIMAGE_REL_I386_DIR32 0x0006
1966 #define MYIMAGE_REL_I386_REL32 0x0014
1969 /* We use myindex to calculate array addresses, rather than
1970 simply doing the normal subscript thing. That's because
1971 some of the above structs have sizes which are not
1972 a whole number of words. GCC rounds their sizes up to a
1973 whole number of words, which means that the address calcs
1974 arising from using normal C indexing or pointer arithmetic
1975 are just plain wrong. Sigh.
1978 myindex ( int scale, void* base, int index )
1981 ((UChar*)base) + scale * index;
1986 printName ( UChar* name, UChar* strtab )
1988 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
1989 UInt32 strtab_offset = * (UInt32*)(name+4);
1990 debugBelch("%s", strtab + strtab_offset );
1993 for (i = 0; i < 8; i++) {
1994 if (name[i] == 0) break;
1995 debugBelch("%c", name[i] );
2002 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
2004 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2005 UInt32 strtab_offset = * (UInt32*)(name+4);
2006 strncpy ( dst, strtab+strtab_offset, dstSize );
2012 if (name[i] == 0) break;
2022 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
2025 /* If the string is longer than 8 bytes, look in the
2026 string table for it -- this will be correctly zero terminated.
2028 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2029 UInt32 strtab_offset = * (UInt32*)(name+4);
2030 return ((UChar*)strtab) + strtab_offset;
2032 /* Otherwise, if shorter than 8 bytes, return the original,
2033 which by defn is correctly terminated.
2035 if (name[7]==0) return name;
2036 /* The annoying case: 8 bytes. Copy into a temporary
2037 (which is never freed ...)
2039 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
2041 strncpy(newstr,name,8);
2047 /* Just compares the short names (first 8 chars) */
2048 static COFF_section *
2049 findPEi386SectionCalled ( ObjectCode* oc, char* name )
2053 = (COFF_header*)(oc->image);
2054 COFF_section* sectab
2056 ((UChar*)(oc->image))
2057 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2059 for (i = 0; i < hdr->NumberOfSections; i++) {
2062 COFF_section* section_i
2064 myindex ( sizeof_COFF_section, sectab, i );
2065 n1 = (UChar*) &(section_i->Name);
2067 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
2068 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
2069 n1[6]==n2[6] && n1[7]==n2[7])
2078 zapTrailingAtSign ( UChar* sym )
2080 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
2082 if (sym[0] == 0) return;
2084 while (sym[i] != 0) i++;
2087 while (j > 0 && my_isdigit(sym[j])) j--;
2088 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
2093 lookupSymbolInDLLs ( UChar *lbl )
2098 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
2099 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
2101 if (lbl[0] == '_') {
2102 /* HACK: if the name has an initial underscore, try stripping
2103 it off & look that up first. I've yet to verify whether there's
2104 a Rule that governs whether an initial '_' *should always* be
2105 stripped off when mapping from import lib name to the DLL name.
2107 sym = GetProcAddress(o_dll->instance, (lbl+1));
2109 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
2113 sym = GetProcAddress(o_dll->instance, lbl);
2115 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
2124 ocVerifyImage_PEi386 ( ObjectCode* oc )
2129 COFF_section* sectab;
2130 COFF_symbol* symtab;
2132 /* debugBelch("\nLOADING %s\n", oc->fileName); */
2133 hdr = (COFF_header*)(oc->image);
2134 sectab = (COFF_section*) (
2135 ((UChar*)(oc->image))
2136 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2138 symtab = (COFF_symbol*) (
2139 ((UChar*)(oc->image))
2140 + hdr->PointerToSymbolTable
2142 strtab = ((UChar*)symtab)
2143 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2145 if (hdr->Machine != 0x14c) {
2146 errorBelch("%s: Not x86 PEi386", oc->fileName);
2149 if (hdr->SizeOfOptionalHeader != 0) {
2150 errorBelch("%s: PEi386 with nonempty optional header", oc->fileName);
2153 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
2154 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
2155 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
2156 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
2157 errorBelch("%s: Not a PEi386 object file", oc->fileName);
2160 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
2161 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
2162 errorBelch("%s: Invalid PEi386 word size or endiannness: %d",
2164 (int)(hdr->Characteristics));
2167 /* If the string table size is way crazy, this might indicate that
2168 there are more than 64k relocations, despite claims to the
2169 contrary. Hence this test. */
2170 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
2172 if ( (*(UInt32*)strtab) > 600000 ) {
2173 /* Note that 600k has no special significance other than being
2174 big enough to handle the almost-2MB-sized lumps that
2175 constitute HSwin32*.o. */
2176 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
2181 /* No further verification after this point; only debug printing. */
2183 IF_DEBUG(linker, i=1);
2184 if (i == 0) return 1;
2186 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
2187 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
2188 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
2191 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
2192 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
2193 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
2194 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
2195 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
2196 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
2197 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
2199 /* Print the section table. */
2201 for (i = 0; i < hdr->NumberOfSections; i++) {
2203 COFF_section* sectab_i
2205 myindex ( sizeof_COFF_section, sectab, i );
2212 printName ( sectab_i->Name, strtab );
2222 sectab_i->VirtualSize,
2223 sectab_i->VirtualAddress,
2224 sectab_i->SizeOfRawData,
2225 sectab_i->PointerToRawData,
2226 sectab_i->NumberOfRelocations,
2227 sectab_i->PointerToRelocations,
2228 sectab_i->PointerToRawData
2230 reltab = (COFF_reloc*) (
2231 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2234 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2235 /* If the relocation field (a short) has overflowed, the
2236 * real count can be found in the first reloc entry.
2238 * See Section 4.1 (last para) of the PE spec (rev6.0).
2240 COFF_reloc* rel = (COFF_reloc*)
2241 myindex ( sizeof_COFF_reloc, reltab, 0 );
2242 noRelocs = rel->VirtualAddress;
2245 noRelocs = sectab_i->NumberOfRelocations;
2249 for (; j < noRelocs; j++) {
2251 COFF_reloc* rel = (COFF_reloc*)
2252 myindex ( sizeof_COFF_reloc, reltab, j );
2254 " type 0x%-4x vaddr 0x%-8x name `",
2256 rel->VirtualAddress );
2257 sym = (COFF_symbol*)
2258 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
2259 /* Hmm..mysterious looking offset - what's it for? SOF */
2260 printName ( sym->Name, strtab -10 );
2267 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
2268 debugBelch("---START of string table---\n");
2269 for (i = 4; i < *(Int32*)strtab; i++) {
2271 debugBelch("\n"); else
2272 debugBelch("%c", strtab[i] );
2274 debugBelch("--- END of string table---\n");
2279 COFF_symbol* symtab_i;
2280 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2281 symtab_i = (COFF_symbol*)
2282 myindex ( sizeof_COFF_symbol, symtab, i );
2288 printName ( symtab_i->Name, strtab );
2297 (Int32)(symtab_i->SectionNumber),
2298 (UInt32)symtab_i->Type,
2299 (UInt32)symtab_i->StorageClass,
2300 (UInt32)symtab_i->NumberOfAuxSymbols
2302 i += symtab_i->NumberOfAuxSymbols;
2312 ocGetNames_PEi386 ( ObjectCode* oc )
2315 COFF_section* sectab;
2316 COFF_symbol* symtab;
2323 hdr = (COFF_header*)(oc->image);
2324 sectab = (COFF_section*) (
2325 ((UChar*)(oc->image))
2326 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2328 symtab = (COFF_symbol*) (
2329 ((UChar*)(oc->image))
2330 + hdr->PointerToSymbolTable
2332 strtab = ((UChar*)(oc->image))
2333 + hdr->PointerToSymbolTable
2334 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2336 /* Allocate space for any (local, anonymous) .bss sections. */
2338 for (i = 0; i < hdr->NumberOfSections; i++) {
2341 COFF_section* sectab_i
2343 myindex ( sizeof_COFF_section, sectab, i );
2344 if (0 != strcmp(sectab_i->Name, ".bss")) continue;
2345 /* sof 10/05: the PE spec text isn't too clear regarding what
2346 * the SizeOfRawData field is supposed to hold for object
2347 * file sections containing just uninitialized data -- for executables,
2348 * it is supposed to be zero; unclear what it's supposed to be
2349 * for object files. However, VirtualSize is guaranteed to be
2350 * zero for object files, which definitely suggests that SizeOfRawData
2351 * will be non-zero (where else would the size of this .bss section be
2352 * stored?) Looking at the COFF_section info for incoming object files,
2353 * this certainly appears to be the case.
2355 * => I suspect we've been incorrectly handling .bss sections in (relocatable)
2356 * object files up until now. This turned out to bite us with ghc-6.4.1's use
2357 * of gcc-3.4.x, which has started to emit initially-zeroed-out local 'static'
2358 * variable decls into to the .bss section. (The specific function in Q which
2359 * triggered this is libraries/base/cbits/dirUtils.c:__hscore_getFolderPath())
2361 if (sectab_i->VirtualSize == 0 && sectab_i->SizeOfRawData == 0) continue;
2362 /* This is a non-empty .bss section. Allocate zeroed space for
2363 it, and set its PointerToRawData field such that oc->image +
2364 PointerToRawData == addr_of_zeroed_space. */
2365 bss_sz = sectab_i->VirtualSize;
2366 if ( bss_sz < sectab_i->SizeOfRawData) { bss_sz = sectab_i->SizeOfRawData; }
2367 zspace = stgCallocBytes(1, bss_sz, "ocGetNames_PEi386(anonymous bss)");
2368 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
2369 addProddableBlock(oc, zspace, bss_sz);
2370 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
2373 /* Copy section information into the ObjectCode. */
2375 for (i = 0; i < hdr->NumberOfSections; i++) {
2381 = SECTIONKIND_OTHER;
2382 COFF_section* sectab_i
2384 myindex ( sizeof_COFF_section, sectab, i );
2385 IF_DEBUG(linker, debugBelch("section name = %s\n", sectab_i->Name ));
2388 /* I'm sure this is the Right Way to do it. However, the
2389 alternative of testing the sectab_i->Name field seems to
2390 work ok with Cygwin.
2392 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
2393 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
2394 kind = SECTIONKIND_CODE_OR_RODATA;
2397 if (0==strcmp(".text",sectab_i->Name) ||
2398 0==strcmp(".rdata",sectab_i->Name)||
2399 0==strcmp(".rodata",sectab_i->Name))
2400 kind = SECTIONKIND_CODE_OR_RODATA;
2401 if (0==strcmp(".data",sectab_i->Name) ||
2402 0==strcmp(".bss",sectab_i->Name))
2403 kind = SECTIONKIND_RWDATA;
2405 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
2406 sz = sectab_i->SizeOfRawData;
2407 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
2409 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
2410 end = start + sz - 1;
2412 if (kind == SECTIONKIND_OTHER
2413 /* Ignore sections called which contain stabs debugging
2415 && 0 != strcmp(".stab", sectab_i->Name)
2416 && 0 != strcmp(".stabstr", sectab_i->Name)
2417 /* ignore constructor section for now */
2418 && 0 != strcmp(".ctors", sectab_i->Name)
2419 /* ignore section generated from .ident */
2420 && 0!= strcmp("/4", sectab_i->Name)
2421 /* ignore unknown section that appeared in gcc 3.4.5(?) */
2422 && 0!= strcmp(".reloc", sectab_i->Name)
2424 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName);
2428 if (kind != SECTIONKIND_OTHER && end >= start) {
2429 addSection(oc, kind, start, end);
2430 addProddableBlock(oc, start, end - start + 1);
2434 /* Copy exported symbols into the ObjectCode. */
2436 oc->n_symbols = hdr->NumberOfSymbols;
2437 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2438 "ocGetNames_PEi386(oc->symbols)");
2439 /* Call me paranoid; I don't care. */
2440 for (i = 0; i < oc->n_symbols; i++)
2441 oc->symbols[i] = NULL;
2445 COFF_symbol* symtab_i;
2446 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2447 symtab_i = (COFF_symbol*)
2448 myindex ( sizeof_COFF_symbol, symtab, i );
2452 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2453 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2454 /* This symbol is global and defined, viz, exported */
2455 /* for MYIMAGE_SYMCLASS_EXTERNAL
2456 && !MYIMAGE_SYM_UNDEFINED,
2457 the address of the symbol is:
2458 address of relevant section + offset in section
2460 COFF_section* sectabent
2461 = (COFF_section*) myindex ( sizeof_COFF_section,
2463 symtab_i->SectionNumber-1 );
2464 addr = ((UChar*)(oc->image))
2465 + (sectabent->PointerToRawData
2469 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2470 && symtab_i->Value > 0) {
2471 /* This symbol isn't in any section at all, ie, global bss.
2472 Allocate zeroed space for it. */
2473 addr = stgCallocBytes(1, symtab_i->Value,
2474 "ocGetNames_PEi386(non-anonymous bss)");
2475 addSection(oc, SECTIONKIND_RWDATA, addr,
2476 ((UChar*)addr) + symtab_i->Value - 1);
2477 addProddableBlock(oc, addr, symtab_i->Value);
2478 /* debugBelch("BSS section at 0x%x\n", addr); */
2481 if (addr != NULL ) {
2482 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2483 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2484 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2485 ASSERT(i >= 0 && i < oc->n_symbols);
2486 /* cstring_from_COFF_symbol_name always succeeds. */
2487 oc->symbols[i] = sname;
2488 ghciInsertStrHashTable(oc->fileName, symhash, sname, addr);
2492 "IGNORING symbol %d\n"
2496 printName ( symtab_i->Name, strtab );
2505 (Int32)(symtab_i->SectionNumber),
2506 (UInt32)symtab_i->Type,
2507 (UInt32)symtab_i->StorageClass,
2508 (UInt32)symtab_i->NumberOfAuxSymbols
2513 i += symtab_i->NumberOfAuxSymbols;
2522 ocResolve_PEi386 ( ObjectCode* oc )
2525 COFF_section* sectab;
2526 COFF_symbol* symtab;
2536 /* ToDo: should be variable-sized? But is at least safe in the
2537 sense of buffer-overrun-proof. */
2539 /* debugBelch("resolving for %s\n", oc->fileName); */
2541 hdr = (COFF_header*)(oc->image);
2542 sectab = (COFF_section*) (
2543 ((UChar*)(oc->image))
2544 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2546 symtab = (COFF_symbol*) (
2547 ((UChar*)(oc->image))
2548 + hdr->PointerToSymbolTable
2550 strtab = ((UChar*)(oc->image))
2551 + hdr->PointerToSymbolTable
2552 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2554 for (i = 0; i < hdr->NumberOfSections; i++) {
2555 COFF_section* sectab_i
2557 myindex ( sizeof_COFF_section, sectab, i );
2560 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2563 /* Ignore sections called which contain stabs debugging
2565 if (0 == strcmp(".stab", sectab_i->Name)
2566 || 0 == strcmp(".stabstr", sectab_i->Name)
2567 || 0 == strcmp(".ctors", sectab_i->Name))
2570 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2571 /* If the relocation field (a short) has overflowed, the
2572 * real count can be found in the first reloc entry.
2574 * See Section 4.1 (last para) of the PE spec (rev6.0).
2576 * Nov2003 update: the GNU linker still doesn't correctly
2577 * handle the generation of relocatable object files with
2578 * overflown relocations. Hence the output to warn of potential
2581 COFF_reloc* rel = (COFF_reloc*)
2582 myindex ( sizeof_COFF_reloc, reltab, 0 );
2583 noRelocs = rel->VirtualAddress;
2585 /* 10/05: we now assume (and check for) a GNU ld that is capable
2586 * of handling object files with (>2^16) of relocs.
2589 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2594 noRelocs = sectab_i->NumberOfRelocations;
2599 for (; j < noRelocs; j++) {
2601 COFF_reloc* reltab_j
2603 myindex ( sizeof_COFF_reloc, reltab, j );
2605 /* the location to patch */
2607 ((UChar*)(oc->image))
2608 + (sectab_i->PointerToRawData
2609 + reltab_j->VirtualAddress
2610 - sectab_i->VirtualAddress )
2612 /* the existing contents of pP */
2614 /* the symbol to connect to */
2615 sym = (COFF_symbol*)
2616 myindex ( sizeof_COFF_symbol,
2617 symtab, reltab_j->SymbolTableIndex );
2620 "reloc sec %2d num %3d: type 0x%-4x "
2621 "vaddr 0x%-8x name `",
2623 (UInt32)reltab_j->Type,
2624 reltab_j->VirtualAddress );
2625 printName ( sym->Name, strtab );
2626 debugBelch("'\n" ));
2628 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2629 COFF_section* section_sym
2630 = findPEi386SectionCalled ( oc, sym->Name );
2632 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2635 S = ((UInt32)(oc->image))
2636 + (section_sym->PointerToRawData
2639 copyName ( sym->Name, strtab, symbol, 1000-1 );
2640 S = (UInt32) lookupSymbol( symbol );
2641 if ((void*)S != NULL) goto foundit;
2642 /* Newline first because the interactive linker has printed "linking..." */
2643 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, symbol);
2647 checkProddableBlock(oc, pP);
2648 switch (reltab_j->Type) {
2649 case MYIMAGE_REL_I386_DIR32:
2652 case MYIMAGE_REL_I386_REL32:
2653 /* Tricky. We have to insert a displacement at
2654 pP which, when added to the PC for the _next_
2655 insn, gives the address of the target (S).
2656 Problem is to know the address of the next insn
2657 when we only know pP. We assume that this
2658 literal field is always the last in the insn,
2659 so that the address of the next insn is pP+4
2660 -- hence the constant 4.
2661 Also I don't know if A should be added, but so
2662 far it has always been zero.
2664 SOF 05/2005: 'A' (old contents of *pP) have been observed
2665 to contain values other than zero (the 'wx' object file
2666 that came with wxhaskell-0.9.4; dunno how it was compiled..).
2667 So, add displacement to old value instead of asserting
2668 A to be zero. Fixes wxhaskell-related crashes, and no other
2669 ill effects have been observed.
2671 Update: the reason why we're seeing these more elaborate
2672 relocations is due to a switch in how the NCG compiles SRTs
2673 and offsets to them from info tables. SRTs live in .(ro)data,
2674 while info tables live in .text, causing GAS to emit REL32/DISP32
2675 relocations with non-zero values. Adding the displacement is
2676 the right thing to do.
2678 *pP = S - ((UInt32)pP) - 4 + A;
2681 debugBelch("%s: unhandled PEi386 relocation type %d",
2682 oc->fileName, reltab_j->Type);
2689 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2693 #endif /* defined(OBJFORMAT_PEi386) */
2696 /* --------------------------------------------------------------------------
2698 * ------------------------------------------------------------------------*/
2700 #if defined(OBJFORMAT_ELF)
2705 #if defined(sparc_HOST_ARCH)
2706 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2707 #elif defined(i386_HOST_ARCH)
2708 # define ELF_TARGET_386 /* Used inside <elf.h> */
2709 #elif defined(x86_64_HOST_ARCH)
2710 # define ELF_TARGET_X64_64
2712 #elif defined (ia64_HOST_ARCH)
2713 # define ELF_TARGET_IA64 /* Used inside <elf.h> */
2715 # define ELF_FUNCTION_DESC /* calling convention uses function descriptors */
2716 # define ELF_NEED_GOT /* needs Global Offset Table */
2717 # define ELF_NEED_PLT /* needs Procedure Linkage Tables */
2720 #if !defined(openbsd_HOST_OS)
2723 /* openbsd elf has things in different places, with diff names */
2724 # include <elf_abi.h>
2725 # include <machine/reloc.h>
2726 # define R_386_32 RELOC_32
2727 # define R_386_PC32 RELOC_PC32
2730 /* If elf.h doesn't define it */
2731 # ifndef R_X86_64_PC64
2732 # define R_X86_64_PC64 24
2736 * Define a set of types which can be used for both ELF32 and ELF64
2740 #define ELFCLASS ELFCLASS64
2741 #define Elf_Addr Elf64_Addr
2742 #define Elf_Word Elf64_Word
2743 #define Elf_Sword Elf64_Sword
2744 #define Elf_Ehdr Elf64_Ehdr
2745 #define Elf_Phdr Elf64_Phdr
2746 #define Elf_Shdr Elf64_Shdr
2747 #define Elf_Sym Elf64_Sym
2748 #define Elf_Rel Elf64_Rel
2749 #define Elf_Rela Elf64_Rela
2750 #define ELF_ST_TYPE ELF64_ST_TYPE
2751 #define ELF_ST_BIND ELF64_ST_BIND
2752 #define ELF_R_TYPE ELF64_R_TYPE
2753 #define ELF_R_SYM ELF64_R_SYM
2755 #define ELFCLASS ELFCLASS32
2756 #define Elf_Addr Elf32_Addr
2757 #define Elf_Word Elf32_Word
2758 #define Elf_Sword Elf32_Sword
2759 #define Elf_Ehdr Elf32_Ehdr
2760 #define Elf_Phdr Elf32_Phdr
2761 #define Elf_Shdr Elf32_Shdr
2762 #define Elf_Sym Elf32_Sym
2763 #define Elf_Rel Elf32_Rel
2764 #define Elf_Rela Elf32_Rela
2766 #define ELF_ST_TYPE ELF32_ST_TYPE
2769 #define ELF_ST_BIND ELF32_ST_BIND
2772 #define ELF_R_TYPE ELF32_R_TYPE
2775 #define ELF_R_SYM ELF32_R_SYM
2781 * Functions to allocate entries in dynamic sections. Currently we simply
2782 * preallocate a large number, and we don't check if a entry for the given
2783 * target already exists (a linear search is too slow). Ideally these
2784 * entries would be associated with symbols.
2787 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
2788 #define GOT_SIZE 0x20000
2789 #define FUNCTION_TABLE_SIZE 0x10000
2790 #define PLT_SIZE 0x08000
2793 static Elf_Addr got[GOT_SIZE];
2794 static unsigned int gotIndex;
2795 static Elf_Addr gp_val = (Elf_Addr)got;
2798 allocateGOTEntry(Elf_Addr target)
2802 if (gotIndex >= GOT_SIZE)
2803 barf("Global offset table overflow");
2805 entry = &got[gotIndex++];
2807 return (Elf_Addr)entry;
2811 #ifdef ELF_FUNCTION_DESC
2817 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
2818 static unsigned int functionTableIndex;
2821 allocateFunctionDesc(Elf_Addr target)
2823 FunctionDesc *entry;
2825 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
2826 barf("Function table overflow");
2828 entry = &functionTable[functionTableIndex++];
2830 entry->gp = (Elf_Addr)gp_val;
2831 return (Elf_Addr)entry;
2835 copyFunctionDesc(Elf_Addr target)
2837 FunctionDesc *olddesc = (FunctionDesc *)target;
2838 FunctionDesc *newdesc;
2840 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
2841 newdesc->gp = olddesc->gp;
2842 return (Elf_Addr)newdesc;
2847 #ifdef ia64_HOST_ARCH
2848 static void ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value);
2849 static void ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc);
2851 static unsigned char plt_code[] =
2853 /* taken from binutils bfd/elfxx-ia64.c */
2854 0x0b, 0x78, 0x00, 0x02, 0x00, 0x24, /* [MMI] addl r15=0,r1;; */
2855 0x00, 0x41, 0x3c, 0x30, 0x28, 0xc0, /* ld8 r16=[r15],8 */
2856 0x01, 0x08, 0x00, 0x84, /* mov r14=r1;; */
2857 0x11, 0x08, 0x00, 0x1e, 0x18, 0x10, /* [MIB] ld8 r1=[r15] */
2858 0x60, 0x80, 0x04, 0x80, 0x03, 0x00, /* mov b6=r16 */
2859 0x60, 0x00, 0x80, 0x00 /* br.few b6;; */
2862 /* If we can't get to the function descriptor via gp, take a local copy of it */
2863 #define PLT_RELOC(code, target) { \
2864 Elf64_Sxword rel_value = target - gp_val; \
2865 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff)) \
2866 ia64_reloc_gprel22((Elf_Addr)code, copyFunctionDesc(target)); \
2868 ia64_reloc_gprel22((Elf_Addr)code, target); \
2873 unsigned char code[sizeof(plt_code)];
2877 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
2879 PLTEntry *plt = (PLTEntry *)oc->plt;
2882 if (oc->pltIndex >= PLT_SIZE)
2883 barf("Procedure table overflow");
2885 entry = &plt[oc->pltIndex++];
2886 memcpy(entry->code, plt_code, sizeof(entry->code));
2887 PLT_RELOC(entry->code, target);
2888 return (Elf_Addr)entry;
2894 return (PLT_SIZE * sizeof(PLTEntry));
2900 * Generic ELF functions
2904 findElfSection ( void* objImage, Elf_Word sh_type )
2906 char* ehdrC = (char*)objImage;
2907 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2908 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
2909 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
2913 for (i = 0; i < ehdr->e_shnum; i++) {
2914 if (shdr[i].sh_type == sh_type
2915 /* Ignore the section header's string table. */
2916 && i != ehdr->e_shstrndx
2917 /* Ignore string tables named .stabstr, as they contain
2919 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
2921 ptr = ehdrC + shdr[i].sh_offset;
2928 #if defined(ia64_HOST_ARCH)
2930 findElfSegment ( void* objImage, Elf_Addr vaddr )
2932 char* ehdrC = (char*)objImage;
2933 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2934 Elf_Phdr* phdr = (Elf_Phdr*)(ehdrC + ehdr->e_phoff);
2935 Elf_Addr segaddr = 0;
2938 for (i = 0; i < ehdr->e_phnum; i++) {
2939 segaddr = phdr[i].p_vaddr;
2940 if ((vaddr >= segaddr) && (vaddr < segaddr + phdr[i].p_memsz))
2948 ocVerifyImage_ELF ( ObjectCode* oc )
2952 int i, j, nent, nstrtab, nsymtabs;
2956 char* ehdrC = (char*)(oc->image);
2957 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
2959 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
2960 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
2961 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
2962 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
2963 errorBelch("%s: not an ELF object", oc->fileName);
2967 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
2968 errorBelch("%s: unsupported ELF format", oc->fileName);
2972 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
2973 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
2975 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
2976 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
2978 errorBelch("%s: unknown endiannness", oc->fileName);
2982 if (ehdr->e_type != ET_REL) {
2983 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
2986 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
2988 IF_DEBUG(linker,debugBelch( "Architecture is " ));
2989 switch (ehdr->e_machine) {
2990 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
2991 #ifdef EM_SPARC32PLUS
2992 case EM_SPARC32PLUS:
2994 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
2996 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
2998 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
3000 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
3001 #elif defined(EM_AMD64)
3002 case EM_AMD64: IF_DEBUG(linker,debugBelch( "amd64" )); break;
3004 default: IF_DEBUG(linker,debugBelch( "unknown" ));
3005 errorBelch("%s: unknown architecture (e_machine == %d)"
3006 , oc->fileName, ehdr->e_machine);
3010 IF_DEBUG(linker,debugBelch(
3011 "\nSection header table: start %ld, n_entries %d, ent_size %d\n",
3012 (long)ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
3014 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
3016 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3018 if (ehdr->e_shstrndx == SHN_UNDEF) {
3019 errorBelch("%s: no section header string table", oc->fileName);
3022 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
3024 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3027 for (i = 0; i < ehdr->e_shnum; i++) {
3028 IF_DEBUG(linker,debugBelch("%2d: ", i ));
3029 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
3030 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
3031 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
3032 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
3033 ehdrC + shdr[i].sh_offset,
3034 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
3036 if (shdr[i].sh_type == SHT_REL) {
3037 IF_DEBUG(linker,debugBelch("Rel " ));
3038 } else if (shdr[i].sh_type == SHT_RELA) {
3039 IF_DEBUG(linker,debugBelch("RelA " ));
3041 IF_DEBUG(linker,debugBelch(" "));
3044 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
3048 IF_DEBUG(linker,debugBelch( "\nString tables" ));
3051 for (i = 0; i < ehdr->e_shnum; i++) {
3052 if (shdr[i].sh_type == SHT_STRTAB
3053 /* Ignore the section header's string table. */
3054 && i != ehdr->e_shstrndx
3055 /* Ignore string tables named .stabstr, as they contain
3057 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3059 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
3060 strtab = ehdrC + shdr[i].sh_offset;
3065 errorBelch("%s: no string tables, or too many", oc->fileName);
3070 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
3071 for (i = 0; i < ehdr->e_shnum; i++) {
3072 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3073 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
3075 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3076 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3077 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%ld rem)\n",
3079 (long)shdr[i].sh_size % sizeof(Elf_Sym)
3081 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
3082 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
3085 for (j = 0; j < nent; j++) {
3086 IF_DEBUG(linker,debugBelch(" %2d ", j ));
3087 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
3088 (int)stab[j].st_shndx,
3089 (int)stab[j].st_size,
3090 (char*)stab[j].st_value ));
3092 IF_DEBUG(linker,debugBelch("type=" ));
3093 switch (ELF_ST_TYPE(stab[j].st_info)) {
3094 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
3095 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
3096 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
3097 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
3098 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
3099 default: IF_DEBUG(linker,debugBelch("? " )); break;
3101 IF_DEBUG(linker,debugBelch(" " ));
3103 IF_DEBUG(linker,debugBelch("bind=" ));
3104 switch (ELF_ST_BIND(stab[j].st_info)) {
3105 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
3106 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
3107 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
3108 default: IF_DEBUG(linker,debugBelch("? " )); break;
3110 IF_DEBUG(linker,debugBelch(" " ));
3112 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
3116 if (nsymtabs == 0) {
3117 errorBelch("%s: didn't find any symbol tables", oc->fileName);
3124 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
3128 if (hdr->sh_type == SHT_PROGBITS
3129 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
3130 /* .text-style section */
3131 return SECTIONKIND_CODE_OR_RODATA;
3134 if (hdr->sh_type == SHT_PROGBITS
3135 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3136 /* .data-style section */
3137 return SECTIONKIND_RWDATA;
3140 if (hdr->sh_type == SHT_PROGBITS
3141 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
3142 /* .rodata-style section */
3143 return SECTIONKIND_CODE_OR_RODATA;
3146 if (hdr->sh_type == SHT_NOBITS
3147 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3148 /* .bss-style section */
3150 return SECTIONKIND_RWDATA;
3153 return SECTIONKIND_OTHER;
3158 ocGetNames_ELF ( ObjectCode* oc )
3163 char* ehdrC = (char*)(oc->image);
3164 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3165 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
3166 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3168 ASSERT(symhash != NULL);
3171 errorBelch("%s: no strtab", oc->fileName);
3176 for (i = 0; i < ehdr->e_shnum; i++) {
3177 /* Figure out what kind of section it is. Logic derived from
3178 Figure 1.14 ("Special Sections") of the ELF document
3179 ("Portable Formats Specification, Version 1.1"). */
3181 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
3183 if (is_bss && shdr[i].sh_size > 0) {
3184 /* This is a non-empty .bss section. Allocate zeroed space for
3185 it, and set its .sh_offset field such that
3186 ehdrC + .sh_offset == addr_of_zeroed_space. */
3187 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
3188 "ocGetNames_ELF(BSS)");
3189 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
3191 debugBelch("BSS section at 0x%x, size %d\n",
3192 zspace, shdr[i].sh_size);
3196 /* fill in the section info */
3197 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
3198 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
3199 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
3200 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
3203 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3205 /* copy stuff into this module's object symbol table */
3206 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3207 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3209 oc->n_symbols = nent;
3210 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3211 "ocGetNames_ELF(oc->symbols)");
3213 for (j = 0; j < nent; j++) {
3215 char isLocal = FALSE; /* avoids uninit-var warning */
3217 char* nm = strtab + stab[j].st_name;
3218 int secno = stab[j].st_shndx;
3220 /* Figure out if we want to add it; if so, set ad to its
3221 address. Otherwise leave ad == NULL. */
3223 if (secno == SHN_COMMON) {
3225 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
3227 debugBelch("COMMON symbol, size %d name %s\n",
3228 stab[j].st_size, nm);
3230 /* Pointless to do addProddableBlock() for this area,
3231 since the linker should never poke around in it. */
3234 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
3235 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
3237 /* and not an undefined symbol */
3238 && stab[j].st_shndx != SHN_UNDEF
3239 /* and not in a "special section" */
3240 && stab[j].st_shndx < SHN_LORESERVE
3242 /* and it's a not a section or string table or anything silly */
3243 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
3244 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
3245 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
3248 /* Section 0 is the undefined section, hence > and not >=. */
3249 ASSERT(secno > 0 && secno < ehdr->e_shnum);
3251 if (shdr[secno].sh_type == SHT_NOBITS) {
3252 debugBelch(" BSS symbol, size %d off %d name %s\n",
3253 stab[j].st_size, stab[j].st_value, nm);
3256 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
3257 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
3260 #ifdef ELF_FUNCTION_DESC
3261 /* dlsym() and the initialisation table both give us function
3262 * descriptors, so to be consistent we store function descriptors
3263 * in the symbol table */
3264 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
3265 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
3267 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s\n",
3268 ad, oc->fileName, nm ));
3273 /* And the decision is ... */
3277 oc->symbols[j] = nm;
3280 /* Ignore entirely. */
3282 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
3286 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
3287 strtab + stab[j].st_name ));
3290 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
3291 (int)ELF_ST_BIND(stab[j].st_info),
3292 (int)ELF_ST_TYPE(stab[j].st_info),
3293 (int)stab[j].st_shndx,
3294 strtab + stab[j].st_name
3297 oc->symbols[j] = NULL;
3306 /* Do ELF relocations which lack an explicit addend. All x86-linux
3307 relocations appear to be of this form. */
3309 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
3310 Elf_Shdr* shdr, int shnum,
3311 Elf_Sym* stab, char* strtab )
3316 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
3317 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
3318 int target_shndx = shdr[shnum].sh_info;
3319 int symtab_shndx = shdr[shnum].sh_link;
3321 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3322 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
3323 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3324 target_shndx, symtab_shndx ));
3326 /* Skip sections that we're not interested in. */
3329 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
3330 if (kind == SECTIONKIND_OTHER) {
3331 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
3336 for (j = 0; j < nent; j++) {
3337 Elf_Addr offset = rtab[j].r_offset;
3338 Elf_Addr info = rtab[j].r_info;
3340 Elf_Addr P = ((Elf_Addr)targ) + offset;
3341 Elf_Word* pP = (Elf_Word*)P;
3346 StgStablePtr stablePtr;
3349 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
3350 j, (void*)offset, (void*)info ));
3352 IF_DEBUG(linker,debugBelch( " ZERO" ));
3355 Elf_Sym sym = stab[ELF_R_SYM(info)];
3356 /* First see if it is a local symbol. */
3357 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3358 /* Yes, so we can get the address directly from the ELF symbol
3360 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3362 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3363 + stab[ELF_R_SYM(info)].st_value);
3366 symbol = strtab + sym.st_name;
3367 stablePtr = (StgStablePtr)lookupHashTable(stablehash, (StgWord)symbol);
3368 if (NULL == stablePtr) {
3369 /* No, so look up the name in our global table. */
3370 S_tmp = lookupSymbol( symbol );
3371 S = (Elf_Addr)S_tmp;
3373 stableVal = deRefStablePtr( stablePtr );
3375 S = (Elf_Addr)S_tmp;
3379 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3382 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
3385 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
3386 (void*)P, (void*)S, (void*)A ));
3387 checkProddableBlock ( oc, pP );
3391 switch (ELF_R_TYPE(info)) {
3392 # ifdef i386_HOST_ARCH
3393 case R_386_32: *pP = value; break;
3394 case R_386_PC32: *pP = value - P; break;
3397 errorBelch("%s: unhandled ELF relocation(Rel) type %lu\n",
3398 oc->fileName, (lnat)ELF_R_TYPE(info));
3406 /* Do ELF relocations for which explicit addends are supplied.
3407 sparc-solaris relocations appear to be of this form. */
3409 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
3410 Elf_Shdr* shdr, int shnum,
3411 Elf_Sym* stab, char* strtab )
3414 char *symbol = NULL;
3416 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
3417 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
3418 int target_shndx = shdr[shnum].sh_info;
3419 int symtab_shndx = shdr[shnum].sh_link;
3421 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3422 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
3423 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3424 target_shndx, symtab_shndx ));
3426 for (j = 0; j < nent; j++) {
3427 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3428 /* This #ifdef only serves to avoid unused-var warnings. */
3429 Elf_Addr offset = rtab[j].r_offset;
3430 Elf_Addr P = targ + offset;
3432 Elf_Addr info = rtab[j].r_info;
3433 Elf_Addr A = rtab[j].r_addend;
3437 # if defined(sparc_HOST_ARCH)
3438 Elf_Word* pP = (Elf_Word*)P;
3440 # elif defined(ia64_HOST_ARCH)
3441 Elf64_Xword *pP = (Elf64_Xword *)P;
3443 # elif defined(powerpc_HOST_ARCH)
3447 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
3448 j, (void*)offset, (void*)info,
3451 IF_DEBUG(linker,debugBelch( " ZERO" ));
3454 Elf_Sym sym = stab[ELF_R_SYM(info)];
3455 /* First see if it is a local symbol. */
3456 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3457 /* Yes, so we can get the address directly from the ELF symbol
3459 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3461 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3462 + stab[ELF_R_SYM(info)].st_value);
3463 #ifdef ELF_FUNCTION_DESC
3464 /* Make a function descriptor for this function */
3465 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
3466 S = allocateFunctionDesc(S + A);
3471 /* No, so look up the name in our global table. */
3472 symbol = strtab + sym.st_name;
3473 S_tmp = lookupSymbol( symbol );
3474 S = (Elf_Addr)S_tmp;
3476 #ifdef ELF_FUNCTION_DESC
3477 /* If a function, already a function descriptor - we would
3478 have to copy it to add an offset. */
3479 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3480 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3484 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3487 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3490 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3491 (void*)P, (void*)S, (void*)A ));
3492 /* checkProddableBlock ( oc, (void*)P ); */
3496 switch (ELF_R_TYPE(info)) {
3497 # if defined(sparc_HOST_ARCH)
3498 case R_SPARC_WDISP30:
3499 w1 = *pP & 0xC0000000;
3500 w2 = (Elf_Word)((value - P) >> 2);
3501 ASSERT((w2 & 0xC0000000) == 0);
3506 w1 = *pP & 0xFFC00000;
3507 w2 = (Elf_Word)(value >> 10);
3508 ASSERT((w2 & 0xFFC00000) == 0);
3514 w2 = (Elf_Word)(value & 0x3FF);
3515 ASSERT((w2 & ~0x3FF) == 0);
3519 /* According to the Sun documentation:
3521 This relocation type resembles R_SPARC_32, except it refers to an
3522 unaligned word. That is, the word to be relocated must be treated
3523 as four separate bytes with arbitrary alignment, not as a word
3524 aligned according to the architecture requirements.
3526 (JRS: which means that freeloading on the R_SPARC_32 case
3527 is probably wrong, but hey ...)
3531 w2 = (Elf_Word)value;
3534 # elif defined(ia64_HOST_ARCH)
3535 case R_IA64_DIR64LSB:
3536 case R_IA64_FPTR64LSB:
3539 case R_IA64_PCREL64LSB:
3542 case R_IA64_SEGREL64LSB:
3543 addr = findElfSegment(ehdrC, value);
3546 case R_IA64_GPREL22:
3547 ia64_reloc_gprel22(P, value);
3549 case R_IA64_LTOFF22:
3550 case R_IA64_LTOFF22X:
3551 case R_IA64_LTOFF_FPTR22:
3552 addr = allocateGOTEntry(value);
3553 ia64_reloc_gprel22(P, addr);
3555 case R_IA64_PCREL21B:
3556 ia64_reloc_pcrel21(P, S, oc);
3559 /* This goes with R_IA64_LTOFF22X and points to the load to
3560 * convert into a move. We don't implement relaxation. */
3562 # elif defined(powerpc_HOST_ARCH)
3563 case R_PPC_ADDR16_LO:
3564 *(Elf32_Half*) P = value;
3567 case R_PPC_ADDR16_HI:
3568 *(Elf32_Half*) P = value >> 16;
3571 case R_PPC_ADDR16_HA:
3572 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3576 *(Elf32_Word *) P = value;
3580 *(Elf32_Word *) P = value - P;
3586 if( delta << 6 >> 6 != delta )
3588 value = (Elf_Addr) (&makeSymbolExtra( oc, ELF_R_SYM(info), value )
3592 if( value == 0 || delta << 6 >> 6 != delta )
3594 barf( "Unable to make SymbolExtra for #%d",
3600 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3601 | (delta & 0x3fffffc);
3605 #if x86_64_HOST_ARCH
3607 *(Elf64_Xword *)P = value;
3612 StgInt64 off = value - P;
3613 if (off >= 0x7fffffffL || off < -0x80000000L) {
3614 #if X86_64_ELF_NONPIC_HACK
3615 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3617 off = pltAddress + A - P;
3619 barf("R_X86_64_PC32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3620 symbol, off, oc->fileName );
3623 *(Elf64_Word *)P = (Elf64_Word)off;
3629 StgInt64 off = value - P;
3630 *(Elf64_Word *)P = (Elf64_Word)off;
3635 if (value >= 0x7fffffffL) {
3636 #if X86_64_ELF_NONPIC_HACK
3637 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3639 value = pltAddress + A;
3641 barf("R_X86_64_32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3642 symbol, value, oc->fileName );
3645 *(Elf64_Word *)P = (Elf64_Word)value;
3649 if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) {
3650 #if X86_64_ELF_NONPIC_HACK
3651 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3653 value = pltAddress + A;
3655 barf("R_X86_64_32S relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3656 symbol, value, oc->fileName );
3659 *(Elf64_Sword *)P = (Elf64_Sword)value;
3662 case R_X86_64_GOTPCREL:
3664 StgInt64 gotAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)->addr;
3665 StgInt64 off = gotAddress + A - P;
3666 *(Elf64_Word *)P = (Elf64_Word)off;
3670 case R_X86_64_PLT32:
3672 StgInt64 off = value - P;
3673 if (off >= 0x7fffffffL || off < -0x80000000L) {
3674 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3676 off = pltAddress + A - P;
3678 *(Elf64_Word *)P = (Elf64_Word)off;
3684 errorBelch("%s: unhandled ELF relocation(RelA) type %lu\n",
3685 oc->fileName, (lnat)ELF_R_TYPE(info));
3694 ocResolve_ELF ( ObjectCode* oc )
3698 Elf_Sym* stab = NULL;
3699 char* ehdrC = (char*)(oc->image);
3700 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3701 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3703 /* first find "the" symbol table */
3704 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3706 /* also go find the string table */
3707 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3709 if (stab == NULL || strtab == NULL) {
3710 errorBelch("%s: can't find string or symbol table", oc->fileName);
3714 /* Process the relocation sections. */
3715 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3716 if (shdr[shnum].sh_type == SHT_REL) {
3717 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3718 shnum, stab, strtab );
3722 if (shdr[shnum].sh_type == SHT_RELA) {
3723 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3724 shnum, stab, strtab );
3729 #if defined(powerpc_HOST_ARCH)
3730 ocFlushInstructionCache( oc );
3738 * Instructions are 41 bits long, packed into 128 bit bundles with a 5-bit template
3739 * at the front. The following utility functions pack and unpack instructions, and
3740 * take care of the most common relocations.
3743 #ifdef ia64_HOST_ARCH
3746 ia64_extract_instruction(Elf64_Xword *target)
3749 int slot = (Elf_Addr)target & 3;
3750 target = (Elf_Addr)target & ~3;
3758 return ((w1 >> 5) & 0x1ffffffffff);
3760 return (w1 >> 46) | ((w2 & 0x7fffff) << 18);
3764 barf("ia64_extract_instruction: invalid slot %p", target);
3769 ia64_deposit_instruction(Elf64_Xword *target, Elf64_Xword value)
3771 int slot = (Elf_Addr)target & 3;
3772 target = (Elf_Addr)target & ~3;
3777 *target |= value << 5;
3780 *target |= value << 46;
3781 *(target+1) |= value >> 18;
3784 *(target+1) |= value << 23;
3790 ia64_reloc_gprel22(Elf_Addr target, Elf_Addr value)
3792 Elf64_Xword instruction;
3793 Elf64_Sxword rel_value;
3795 rel_value = value - gp_val;
3796 if ((rel_value > 0x1fffff) || (rel_value < -0x1fffff))
3797 barf("GP-relative data out of range (address = 0x%lx, gp = 0x%lx)", value, gp_val);
3799 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3800 instruction |= (((rel_value >> 0) & 0x07f) << 13) /* imm7b */
3801 | (((rel_value >> 7) & 0x1ff) << 27) /* imm9d */
3802 | (((rel_value >> 16) & 0x01f) << 22) /* imm5c */
3803 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3804 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3808 ia64_reloc_pcrel21(Elf_Addr target, Elf_Addr value, ObjectCode *oc)
3810 Elf64_Xword instruction;
3811 Elf64_Sxword rel_value;
3814 entry = allocatePLTEntry(value, oc);
3816 rel_value = (entry >> 4) - (target >> 4);
3817 if ((rel_value > 0xfffff) || (rel_value < -0xfffff))
3818 barf("PLT entry too far away (entry = 0x%lx, target = 0x%lx)", entry, target);
3820 instruction = ia64_extract_instruction((Elf64_Xword *)target);
3821 instruction |= ((rel_value & 0xfffff) << 13) /* imm20b */
3822 | ((Elf64_Xword)(rel_value < 0) << 36); /* s */
3823 ia64_deposit_instruction((Elf64_Xword *)target, instruction);
3829 * PowerPC & X86_64 ELF specifics
3832 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3834 static int ocAllocateSymbolExtras_ELF( ObjectCode *oc )
3840 ehdr = (Elf_Ehdr *) oc->image;
3841 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3843 for( i = 0; i < ehdr->e_shnum; i++ )
3844 if( shdr[i].sh_type == SHT_SYMTAB )
3847 if( i == ehdr->e_shnum )
3849 errorBelch( "This ELF file contains no symtab" );
3853 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3855 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3856 (int) shdr[i].sh_entsize, (int) sizeof( Elf_Sym ) );
3861 return ocAllocateSymbolExtras( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3864 #endif /* powerpc */
3868 /* --------------------------------------------------------------------------
3870 * ------------------------------------------------------------------------*/
3872 #if defined(OBJFORMAT_MACHO)
3875 Support for MachO linking on Darwin/MacOS X
3876 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3878 I hereby formally apologize for the hackish nature of this code.
3879 Things that need to be done:
3880 *) implement ocVerifyImage_MachO
3881 *) add still more sanity checks.
3884 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
3885 #define mach_header mach_header_64
3886 #define segment_command segment_command_64
3887 #define section section_64
3888 #define nlist nlist_64
3891 #ifdef powerpc_HOST_ARCH
3892 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3894 struct mach_header *header = (struct mach_header *) oc->image;
3895 struct load_command *lc = (struct load_command *) (header + 1);
3898 for( i = 0; i < header->ncmds; i++ )
3900 if( lc->cmd == LC_SYMTAB )
3902 // Find out the first and last undefined external
3903 // symbol, so we don't have to allocate too many
3905 struct symtab_command *symLC = (struct symtab_command *) lc;
3906 unsigned min = symLC->nsyms, max = 0;
3907 struct nlist *nlist =
3908 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
3910 for(i=0;i<symLC->nsyms;i++)
3912 if(nlist[i].n_type & N_STAB)
3914 else if(nlist[i].n_type & N_EXT)
3916 if((nlist[i].n_type & N_TYPE) == N_UNDF
3917 && (nlist[i].n_value == 0))
3927 return ocAllocateSymbolExtras(oc, max - min + 1, min);
3932 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3934 return ocAllocateSymbolExtras(oc,0,0);
3937 #ifdef x86_64_HOST_ARCH
3938 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3940 struct mach_header *header = (struct mach_header *) oc->image;
3941 struct load_command *lc = (struct load_command *) (header + 1);
3944 for( i = 0; i < header->ncmds; i++ )
3946 if( lc->cmd == LC_SYMTAB )
3948 // Just allocate one entry for every symbol
3949 struct symtab_command *symLC = (struct symtab_command *) lc;
3951 return ocAllocateSymbolExtras(oc, symLC->nsyms, 0);
3954 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3956 return ocAllocateSymbolExtras(oc,0,0);
3960 static int ocVerifyImage_MachO(ObjectCode* oc)
3962 char *image = (char*) oc->image;
3963 struct mach_header *header = (struct mach_header*) image;
3965 #if x86_64_TARGET_ARCH || powerpc64_TARGET_ARCH
3966 if(header->magic != MH_MAGIC_64)
3969 if(header->magic != MH_MAGIC)
3972 // FIXME: do some more verifying here
3976 static int resolveImports(
3979 struct symtab_command *symLC,
3980 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
3981 unsigned long *indirectSyms,
3982 struct nlist *nlist)
3985 size_t itemSize = 4;
3988 int isJumpTable = 0;
3989 if(!strcmp(sect->sectname,"__jump_table"))
3993 ASSERT(sect->reserved2 == itemSize);
3997 for(i=0; i*itemSize < sect->size;i++)
3999 // according to otool, reserved1 contains the first index into the indirect symbol table
4000 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
4001 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4004 if((symbol->n_type & N_TYPE) == N_UNDF
4005 && (symbol->n_type & N_EXT) && (symbol->n_value != 0))
4006 addr = (void*) (symbol->n_value);
4008 addr = lookupSymbol(nm);
4011 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
4019 checkProddableBlock(oc,image + sect->offset + i*itemSize);
4020 *(image + sect->offset + i*itemSize) = 0xe9; // jmp
4021 *(unsigned*)(image + sect->offset + i*itemSize + 1)
4022 = (char*)addr - (image + sect->offset + i*itemSize + 5);
4027 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
4028 ((void**)(image + sect->offset))[i] = addr;
4035 static unsigned long relocateAddress(
4038 struct section* sections,
4039 unsigned long address)
4042 for(i = 0; i < nSections; i++)
4044 if(sections[i].addr <= address
4045 && address < sections[i].addr + sections[i].size)
4047 return (unsigned long)oc->image
4048 + sections[i].offset + address - sections[i].addr;
4051 barf("Invalid Mach-O file:"
4052 "Address out of bounds while relocating object file");
4056 static int relocateSection(
4059 struct symtab_command *symLC, struct nlist *nlist,
4060 int nSections, struct section* sections, struct section *sect)
4062 struct relocation_info *relocs;
4065 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
4067 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
4069 else if(!strcmp(sect->sectname,"__la_sym_ptr2"))
4071 else if(!strcmp(sect->sectname,"__la_sym_ptr3"))
4075 relocs = (struct relocation_info*) (image + sect->reloff);
4079 #ifdef x86_64_HOST_ARCH
4080 struct relocation_info *reloc = &relocs[i];
4082 char *thingPtr = image + sect->offset + reloc->r_address;
4086 int type = reloc->r_type;
4088 checkProddableBlock(oc,thingPtr);
4089 switch(reloc->r_length)
4092 thing = *(uint8_t*)thingPtr;
4093 baseValue = (uint64_t)thingPtr + 1;
4096 thing = *(uint16_t*)thingPtr;
4097 baseValue = (uint64_t)thingPtr + 2;
4100 thing = *(uint32_t*)thingPtr;
4101 baseValue = (uint64_t)thingPtr + 4;
4104 thing = *(uint64_t*)thingPtr;
4105 baseValue = (uint64_t)thingPtr + 8;
4108 barf("Unknown size.");
4111 if(type == X86_64_RELOC_GOT
4112 || type == X86_64_RELOC_GOT_LOAD)
4114 ASSERT(reloc->r_extern);
4115 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)->addr;
4117 type = X86_64_RELOC_SIGNED;
4119 else if(reloc->r_extern)
4121 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4122 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4123 if(symbol->n_value == 0)
4124 value = (uint64_t) lookupSymbol(nm);
4126 value = relocateAddress(oc, nSections, sections,
4131 value = sections[reloc->r_symbolnum-1].offset
4132 - sections[reloc->r_symbolnum-1].addr
4136 if(type == X86_64_RELOC_BRANCH)
4138 if((int32_t)(value - baseValue) != (int64_t)(value - baseValue))
4140 ASSERT(reloc->r_extern);
4141 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)
4144 ASSERT((int32_t)(value - baseValue) == (int64_t)(value - baseValue));
4145 type = X86_64_RELOC_SIGNED;
4150 case X86_64_RELOC_UNSIGNED:
4151 ASSERT(!reloc->r_pcrel);
4154 case X86_64_RELOC_SIGNED:
4155 ASSERT(reloc->r_pcrel);
4156 thing += value - baseValue;
4158 case X86_64_RELOC_SUBTRACTOR:
4159 ASSERT(!reloc->r_pcrel);
4163 barf("unkown relocation");
4166 switch(reloc->r_length)
4169 *(uint8_t*)thingPtr = thing;
4172 *(uint16_t*)thingPtr = thing;
4175 *(uint32_t*)thingPtr = thing;
4178 *(uint64_t*)thingPtr = thing;
4182 if(relocs[i].r_address & R_SCATTERED)
4184 struct scattered_relocation_info *scat =
4185 (struct scattered_relocation_info*) &relocs[i];
4189 if(scat->r_length == 2)
4191 unsigned long word = 0;
4192 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
4193 checkProddableBlock(oc,wordPtr);
4195 // Note on relocation types:
4196 // i386 uses the GENERIC_RELOC_* types,
4197 // while ppc uses special PPC_RELOC_* types.
4198 // *_RELOC_VANILLA and *_RELOC_PAIR have the same value
4199 // in both cases, all others are different.
4200 // Therefore, we use GENERIC_RELOC_VANILLA
4201 // and GENERIC_RELOC_PAIR instead of the PPC variants,
4202 // and use #ifdefs for the other types.
4204 // Step 1: Figure out what the relocated value should be
4205 if(scat->r_type == GENERIC_RELOC_VANILLA)
4207 word = *wordPtr + (unsigned long) relocateAddress(
4214 #ifdef powerpc_HOST_ARCH
4215 else if(scat->r_type == PPC_RELOC_SECTDIFF
4216 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
4217 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
4218 || scat->r_type == PPC_RELOC_HA16_SECTDIFF)
4220 else if(scat->r_type == GENERIC_RELOC_SECTDIFF)
4223 struct scattered_relocation_info *pair =
4224 (struct scattered_relocation_info*) &relocs[i+1];
4226 if(!pair->r_scattered || pair->r_type != GENERIC_RELOC_PAIR)
4227 barf("Invalid Mach-O file: "
4228 "RELOC_*_SECTDIFF not followed by RELOC_PAIR");
4230 word = (unsigned long)
4231 (relocateAddress(oc, nSections, sections, scat->r_value)
4232 - relocateAddress(oc, nSections, sections, pair->r_value));
4235 #ifdef powerpc_HOST_ARCH
4236 else if(scat->r_type == PPC_RELOC_HI16
4237 || scat->r_type == PPC_RELOC_LO16
4238 || scat->r_type == PPC_RELOC_HA16
4239 || scat->r_type == PPC_RELOC_LO14)
4240 { // these are generated by label+offset things
4241 struct relocation_info *pair = &relocs[i+1];
4242 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
4243 barf("Invalid Mach-O file: "
4244 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
4246 if(scat->r_type == PPC_RELOC_LO16)
4248 word = ((unsigned short*) wordPtr)[1];
4249 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4251 else if(scat->r_type == PPC_RELOC_LO14)
4253 barf("Unsupported Relocation: PPC_RELOC_LO14");
4254 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
4255 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4257 else if(scat->r_type == PPC_RELOC_HI16)
4259 word = ((unsigned short*) wordPtr)[1] << 16;
4260 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4262 else if(scat->r_type == PPC_RELOC_HA16)
4264 word = ((unsigned short*) wordPtr)[1] << 16;
4265 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4269 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
4276 continue; // ignore the others
4278 #ifdef powerpc_HOST_ARCH
4279 if(scat->r_type == GENERIC_RELOC_VANILLA
4280 || scat->r_type == PPC_RELOC_SECTDIFF)
4282 if(scat->r_type == GENERIC_RELOC_VANILLA
4283 || scat->r_type == GENERIC_RELOC_SECTDIFF)
4288 #ifdef powerpc_HOST_ARCH
4289 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
4291 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4293 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
4295 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4297 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
4299 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4300 + ((word & (1<<15)) ? 1 : 0);
4306 continue; // FIXME: I hope it's OK to ignore all the others.
4310 struct relocation_info *reloc = &relocs[i];
4311 if(reloc->r_pcrel && !reloc->r_extern)
4314 if(reloc->r_length == 2)
4316 unsigned long word = 0;
4317 #ifdef powerpc_HOST_ARCH
4318 unsigned long jumpIsland = 0;
4319 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
4320 // to avoid warning and to catch
4324 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
4325 checkProddableBlock(oc,wordPtr);
4327 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4331 #ifdef powerpc_HOST_ARCH
4332 else if(reloc->r_type == PPC_RELOC_LO16)
4334 word = ((unsigned short*) wordPtr)[1];
4335 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4337 else if(reloc->r_type == PPC_RELOC_HI16)
4339 word = ((unsigned short*) wordPtr)[1] << 16;
4340 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4342 else if(reloc->r_type == PPC_RELOC_HA16)
4344 word = ((unsigned short*) wordPtr)[1] << 16;
4345 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4347 else if(reloc->r_type == PPC_RELOC_BR24)
4350 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
4354 if(!reloc->r_extern)
4357 sections[reloc->r_symbolnum-1].offset
4358 - sections[reloc->r_symbolnum-1].addr
4365 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4366 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4367 void *symbolAddress = lookupSymbol(nm);
4370 errorBelch("\nunknown symbol `%s'", nm);
4376 #ifdef powerpc_HOST_ARCH
4377 // In the .o file, this should be a relative jump to NULL
4378 // and we'll change it to a relative jump to the symbol
4379 ASSERT(word + reloc->r_address == 0);
4380 jumpIsland = (unsigned long)
4381 &makeSymbolExtra(oc,
4383 (unsigned long) symbolAddress)
4387 offsetToJumpIsland = word + jumpIsland
4388 - (((long)image) + sect->offset - sect->addr);
4391 word += (unsigned long) symbolAddress
4392 - (((long)image) + sect->offset - sect->addr);
4396 word += (unsigned long) symbolAddress;
4400 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4405 #ifdef powerpc_HOST_ARCH
4406 else if(reloc->r_type == PPC_RELOC_LO16)
4408 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4411 else if(reloc->r_type == PPC_RELOC_HI16)
4413 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4416 else if(reloc->r_type == PPC_RELOC_HA16)
4418 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4419 + ((word & (1<<15)) ? 1 : 0);
4422 else if(reloc->r_type == PPC_RELOC_BR24)
4424 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4426 // The branch offset is too large.
4427 // Therefore, we try to use a jump island.
4430 barf("unconditional relative branch out of range: "
4431 "no jump island available");
4434 word = offsetToJumpIsland;
4435 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4436 barf("unconditional relative branch out of range: "
4437 "jump island out of range");
4439 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
4444 barf("\nunknown relocation %d",reloc->r_type);
4452 static int ocGetNames_MachO(ObjectCode* oc)
4454 char *image = (char*) oc->image;
4455 struct mach_header *header = (struct mach_header*) image;
4456 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4457 unsigned i,curSymbol = 0;
4458 struct segment_command *segLC = NULL;
4459 struct section *sections;
4460 struct symtab_command *symLC = NULL;
4461 struct nlist *nlist;
4462 unsigned long commonSize = 0;
4463 char *commonStorage = NULL;
4464 unsigned long commonCounter;
4466 for(i=0;i<header->ncmds;i++)
4468 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4469 segLC = (struct segment_command*) lc;
4470 else if(lc->cmd == LC_SYMTAB)
4471 symLC = (struct symtab_command*) lc;
4472 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4475 sections = (struct section*) (segLC+1);
4476 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4480 barf("ocGetNames_MachO: no segment load command");
4482 for(i=0;i<segLC->nsects;i++)
4484 if(sections[i].size == 0)
4487 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
4489 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
4490 "ocGetNames_MachO(common symbols)");
4491 sections[i].offset = zeroFillArea - image;
4494 if(!strcmp(sections[i].sectname,"__text"))
4495 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
4496 (void*) (image + sections[i].offset),
4497 (void*) (image + sections[i].offset + sections[i].size));
4498 else if(!strcmp(sections[i].sectname,"__const"))
4499 addSection(oc, SECTIONKIND_RWDATA,
4500 (void*) (image + sections[i].offset),
4501 (void*) (image + sections[i].offset + sections[i].size));
4502 else if(!strcmp(sections[i].sectname,"__data"))
4503 addSection(oc, SECTIONKIND_RWDATA,
4504 (void*) (image + sections[i].offset),
4505 (void*) (image + sections[i].offset + sections[i].size));
4506 else if(!strcmp(sections[i].sectname,"__bss")
4507 || !strcmp(sections[i].sectname,"__common"))
4508 addSection(oc, SECTIONKIND_RWDATA,
4509 (void*) (image + sections[i].offset),
4510 (void*) (image + sections[i].offset + sections[i].size));
4512 addProddableBlock(oc, (void*) (image + sections[i].offset),
4516 // count external symbols defined here
4520 for(i=0;i<symLC->nsyms;i++)
4522 if(nlist[i].n_type & N_STAB)
4524 else if(nlist[i].n_type & N_EXT)
4526 if((nlist[i].n_type & N_TYPE) == N_UNDF
4527 && (nlist[i].n_value != 0))
4529 commonSize += nlist[i].n_value;
4532 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4537 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
4538 "ocGetNames_MachO(oc->symbols)");
4542 for(i=0;i<symLC->nsyms;i++)
4544 if(nlist[i].n_type & N_STAB)
4546 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4548 if(nlist[i].n_type & N_EXT)
4550 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4551 if((nlist[i].n_desc & N_WEAK_DEF) && lookupSymbol(nm))
4552 ; // weak definition, and we already have a definition
4555 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4557 + sections[nlist[i].n_sect-1].offset
4558 - sections[nlist[i].n_sect-1].addr
4559 + nlist[i].n_value);
4560 oc->symbols[curSymbol++] = nm;
4567 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
4568 commonCounter = (unsigned long)commonStorage;
4571 for(i=0;i<symLC->nsyms;i++)
4573 if((nlist[i].n_type & N_TYPE) == N_UNDF
4574 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
4576 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4577 unsigned long sz = nlist[i].n_value;
4579 nlist[i].n_value = commonCounter;
4581 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4582 (void*)commonCounter);
4583 oc->symbols[curSymbol++] = nm;
4585 commonCounter += sz;
4592 static int ocResolve_MachO(ObjectCode* oc)
4594 char *image = (char*) oc->image;
4595 struct mach_header *header = (struct mach_header*) image;
4596 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4598 struct segment_command *segLC = NULL;
4599 struct section *sections;
4600 struct symtab_command *symLC = NULL;
4601 struct dysymtab_command *dsymLC = NULL;
4602 struct nlist *nlist;
4604 for(i=0;i<header->ncmds;i++)
4606 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4607 segLC = (struct segment_command*) lc;
4608 else if(lc->cmd == LC_SYMTAB)
4609 symLC = (struct symtab_command*) lc;
4610 else if(lc->cmd == LC_DYSYMTAB)
4611 dsymLC = (struct dysymtab_command*) lc;
4612 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4615 sections = (struct section*) (segLC+1);
4616 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4621 unsigned long *indirectSyms
4622 = (unsigned long*) (image + dsymLC->indirectsymoff);
4624 for(i=0;i<segLC->nsects;i++)
4626 if( !strcmp(sections[i].sectname,"__la_symbol_ptr")
4627 || !strcmp(sections[i].sectname,"__la_sym_ptr2")
4628 || !strcmp(sections[i].sectname,"__la_sym_ptr3"))
4630 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4633 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr")
4634 || !strcmp(sections[i].sectname,"__pointers"))
4636 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4639 else if(!strcmp(sections[i].sectname,"__jump_table"))
4641 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4647 for(i=0;i<segLC->nsects;i++)
4649 if(!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
4653 #if defined (powerpc_HOST_ARCH)
4654 ocFlushInstructionCache( oc );
4660 #ifdef powerpc_HOST_ARCH
4662 * The Mach-O object format uses leading underscores. But not everywhere.
4663 * There is a small number of runtime support functions defined in
4664 * libcc_dynamic.a whose name does not have a leading underscore.
4665 * As a consequence, we can't get their address from C code.
4666 * We have to use inline assembler just to take the address of a function.
4670 static void machoInitSymbolsWithoutUnderscore()
4672 extern void* symbolsWithoutUnderscore[];
4673 void **p = symbolsWithoutUnderscore;
4674 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
4677 #define SymI_NeedsProto(x) \
4678 __asm__ volatile(".long " # x);
4680 RTS_MACHO_NOUNDERLINE_SYMBOLS
4682 __asm__ volatile(".text");
4685 #define SymI_NeedsProto(x) \
4686 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
4688 RTS_MACHO_NOUNDERLINE_SYMBOLS
4695 * Figure out by how much to shift the entire Mach-O file in memory
4696 * when loading so that its single segment ends up 16-byte-aligned
4698 static int machoGetMisalignment( FILE * f )
4700 struct mach_header header;
4703 fread(&header, sizeof(header), 1, f);
4706 #if x86_64_TARGET_ARCH || powerpc64_TARGET_ARCH
4707 if(header.magic != MH_MAGIC_64)
4710 if(header.magic != MH_MAGIC)
4714 misalignment = (header.sizeofcmds + sizeof(header))
4717 return misalignment ? (16 - misalignment) : 0;