add numSparks# primop (#4167)
[ghc.git] / rts / Linker.c
1 /* -----------------------------------------------------------------------------
2 *
3 * (c) The GHC Team, 2000-2004
4 *
5 * RTS Object Linker
6 *
7 * ---------------------------------------------------------------------------*/
8
9 #if 0
10 #include "PosixSource.h"
11 #endif
12
13 /* Linux needs _GNU_SOURCE to get RTLD_DEFAULT from <dlfcn.h> and
14 MREMAP_MAYMOVE from <sys/mman.h>.
15 */
16 #ifdef __linux__
17 #define _GNU_SOURCE
18 #endif
19
20 #include "Rts.h"
21 #include "HsFFI.h"
22
23 #include "sm/Storage.h"
24 #include "Stats.h"
25 #include "Hash.h"
26 #include "LinkerInternals.h"
27 #include "RtsUtils.h"
28 #include "Trace.h"
29 #include "StgPrimFloat.h" // for __int_encodeFloat etc.
30 #include "Stable.h"
31
32 #if !defined(mingw32_HOST_OS)
33 #include "posix/Signals.h"
34 #endif
35
36 #if defined(mingw32_HOST_OS)
37 // get protos for is*()
38 #include <ctype.h>
39 #endif
40
41 #ifdef HAVE_SYS_TYPES_H
42 #include <sys/types.h>
43 #endif
44
45 #include <stdlib.h>
46 #include <string.h>
47 #include <stdio.h>
48 #include <assert.h>
49
50 #ifdef HAVE_SYS_STAT_H
51 #include <sys/stat.h>
52 #endif
53
54 #if defined(HAVE_DLFCN_H)
55 #include <dlfcn.h>
56 #endif
57
58 #if defined(cygwin32_HOST_OS)
59 #ifdef HAVE_DIRENT_H
60 #include <dirent.h>
61 #endif
62
63 #ifdef HAVE_SYS_TIME_H
64 #include <sys/time.h>
65 #endif
66 #include <regex.h>
67 #include <sys/fcntl.h>
68 #include <sys/termios.h>
69 #include <sys/utime.h>
70 #include <sys/utsname.h>
71 #include <sys/wait.h>
72 #endif
73
74 #if defined(linux_HOST_OS) || defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS) || defined(darwin_HOST_OS)
75 #define USE_MMAP
76 #include <fcntl.h>
77 #include <sys/mman.h>
78
79 #ifdef HAVE_UNISTD_H
80 #include <unistd.h>
81 #endif
82
83 #endif
84
85 #if defined(linux_HOST_OS) || defined(solaris2_HOST_OS) || defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS)
86 # define OBJFORMAT_ELF
87 # include <regex.h> // regex is already used by dlopen() so this is OK
88 // to use here without requiring an additional lib
89 #elif defined(cygwin32_HOST_OS) || defined (mingw32_HOST_OS)
90 # define OBJFORMAT_PEi386
91 # include <windows.h>
92 # include <math.h>
93 #elif defined(darwin_HOST_OS)
94 # define OBJFORMAT_MACHO
95 # include <regex.h>
96 # include <mach-o/loader.h>
97 # include <mach-o/nlist.h>
98 # include <mach-o/reloc.h>
99 #if !defined(HAVE_DLFCN_H)
100 # include <mach-o/dyld.h>
101 #endif
102 #if defined(powerpc_HOST_ARCH)
103 # include <mach-o/ppc/reloc.h>
104 #endif
105 #if defined(x86_64_HOST_ARCH)
106 # include <mach-o/x86_64/reloc.h>
107 #endif
108 #endif
109
110 /* Hash table mapping symbol names to Symbol */
111 static /*Str*/HashTable *symhash;
112
113 /* Hash table mapping symbol names to StgStablePtr */
114 static /*Str*/HashTable *stablehash;
115
116 /* List of currently loaded objects */
117 ObjectCode *objects = NULL; /* initially empty */
118
119 #if defined(OBJFORMAT_ELF)
120 static int ocVerifyImage_ELF ( ObjectCode* oc );
121 static int ocGetNames_ELF ( ObjectCode* oc );
122 static int ocResolve_ELF ( ObjectCode* oc );
123 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
124 static int ocAllocateSymbolExtras_ELF ( ObjectCode* oc );
125 #endif
126 #elif defined(OBJFORMAT_PEi386)
127 static int ocVerifyImage_PEi386 ( ObjectCode* oc );
128 static int ocGetNames_PEi386 ( ObjectCode* oc );
129 static int ocResolve_PEi386 ( ObjectCode* oc );
130 static void *lookupSymbolInDLLs ( unsigned char *lbl );
131 static void zapTrailingAtSign ( unsigned char *sym );
132 #elif defined(OBJFORMAT_MACHO)
133 static int ocVerifyImage_MachO ( ObjectCode* oc );
134 static int ocGetNames_MachO ( ObjectCode* oc );
135 static int ocResolve_MachO ( ObjectCode* oc );
136
137 #ifndef USE_MMAP
138 static int machoGetMisalignment( FILE * );
139 #endif
140 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
141 static int ocAllocateSymbolExtras_MachO ( ObjectCode* oc );
142 #endif
143 #ifdef powerpc_HOST_ARCH
144 static void machoInitSymbolsWithoutUnderscore( void );
145 #endif
146 #endif
147
148 /* on x86_64 we have a problem with relocating symbol references in
149 * code that was compiled without -fPIC. By default, the small memory
150 * model is used, which assumes that symbol references can fit in a
151 * 32-bit slot. The system dynamic linker makes this work for
152 * references to shared libraries by either (a) allocating a jump
153 * table slot for code references, or (b) moving the symbol at load
154 * time (and copying its contents, if necessary) for data references.
155 *
156 * We unfortunately can't tell whether symbol references are to code
157 * or data. So for now we assume they are code (the vast majority
158 * are), and allocate jump-table slots. Unfortunately this will
159 * SILENTLY generate crashing code for data references. This hack is
160 * enabled by X86_64_ELF_NONPIC_HACK.
161 *
162 * One workaround is to use shared Haskell libraries. This is
163 * coming. Another workaround is to keep the static libraries but
164 * compile them with -fPIC, because that will generate PIC references
165 * to data which can be relocated. The PIC code is still too green to
166 * do this systematically, though.
167 *
168 * See bug #781
169 * See thread http://www.haskell.org/pipermail/cvs-ghc/2007-September/038458.html
170 *
171 * Naming Scheme for Symbol Macros
172 *
173 * SymI_*: symbol is internal to the RTS. It resides in an object
174 * file/library that is statically.
175 * SymE_*: symbol is external to the RTS library. It might be linked
176 * dynamically.
177 *
178 * Sym*_HasProto : the symbol prototype is imported in an include file
179 * or defined explicitly
180 * Sym*_NeedsProto: the symbol is undefined and we add a dummy
181 * default proto extern void sym(void);
182 */
183 #define X86_64_ELF_NONPIC_HACK 1
184
185 /* Link objects into the lower 2Gb on x86_64. GHC assumes the
186 * small memory model on this architecture (see gcc docs,
187 * -mcmodel=small).
188 *
189 * MAP_32BIT not available on OpenBSD/amd64
190 */
191 #if defined(x86_64_HOST_ARCH) && defined(MAP_32BIT)
192 #define TRY_MAP_32BIT MAP_32BIT
193 #else
194 #define TRY_MAP_32BIT 0
195 #endif
196
197 /*
198 * Due to the small memory model (see above), on x86_64 we have to map
199 * all our non-PIC object files into the low 2Gb of the address space
200 * (why 2Gb and not 4Gb? Because all addresses must be reachable
201 * using a 32-bit signed PC-relative offset). On Linux we can do this
202 * using the MAP_32BIT flag to mmap(), however on other OSs
203 * (e.g. *BSD, see #2063, and also on Linux inside Xen, see #2512), we
204 * can't do this. So on these systems, we have to pick a base address
205 * in the low 2Gb of the address space and try to allocate memory from
206 * there.
207 *
208 * We pick a default address based on the OS, but also make this
209 * configurable via an RTS flag (+RTS -xm)
210 */
211 #if defined(x86_64_HOST_ARCH)
212
213 #if defined(MAP_32BIT)
214 // Try to use MAP_32BIT
215 #define MMAP_32BIT_BASE_DEFAULT 0
216 #else
217 // A guess: 1Gb.
218 #define MMAP_32BIT_BASE_DEFAULT 0x40000000
219 #endif
220
221 static void *mmap_32bit_base = (void *)MMAP_32BIT_BASE_DEFAULT;
222 #endif
223
224 /* MAP_ANONYMOUS is MAP_ANON on some systems, e.g. OpenBSD */
225 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
226 #define MAP_ANONYMOUS MAP_ANON
227 #endif
228
229 /* -----------------------------------------------------------------------------
230 * Built-in symbols from the RTS
231 */
232
233 typedef struct _RtsSymbolVal {
234 char *lbl;
235 void *addr;
236 } RtsSymbolVal;
237
238 #define Maybe_Stable_Names SymI_HasProto(stg_mkWeakzh) \
239 SymI_HasProto(stg_mkWeakForeignEnvzh) \
240 SymI_HasProto(stg_makeStableNamezh) \
241 SymI_HasProto(stg_finalizzeWeakzh)
242
243 #if !defined (mingw32_HOST_OS)
244 #define RTS_POSIX_ONLY_SYMBOLS \
245 SymI_HasProto(__hscore_get_saved_termios) \
246 SymI_HasProto(__hscore_set_saved_termios) \
247 SymI_HasProto(shutdownHaskellAndSignal) \
248 SymI_HasProto(lockFile) \
249 SymI_HasProto(unlockFile) \
250 SymI_HasProto(signal_handlers) \
251 SymI_HasProto(stg_sig_install) \
252 SymI_NeedsProto(nocldstop)
253 #endif
254
255 #if defined (cygwin32_HOST_OS)
256 #define RTS_MINGW_ONLY_SYMBOLS /**/
257 /* Don't have the ability to read import libs / archives, so
258 * we have to stupidly list a lot of what libcygwin.a
259 * exports; sigh.
260 */
261 #define RTS_CYGWIN_ONLY_SYMBOLS \
262 SymI_HasProto(regfree) \
263 SymI_HasProto(regexec) \
264 SymI_HasProto(regerror) \
265 SymI_HasProto(regcomp) \
266 SymI_HasProto(__errno) \
267 SymI_HasProto(access) \
268 SymI_HasProto(chmod) \
269 SymI_HasProto(chdir) \
270 SymI_HasProto(close) \
271 SymI_HasProto(creat) \
272 SymI_HasProto(dup) \
273 SymI_HasProto(dup2) \
274 SymI_HasProto(fstat) \
275 SymI_HasProto(fcntl) \
276 SymI_HasProto(getcwd) \
277 SymI_HasProto(getenv) \
278 SymI_HasProto(lseek) \
279 SymI_HasProto(open) \
280 SymI_HasProto(fpathconf) \
281 SymI_HasProto(pathconf) \
282 SymI_HasProto(stat) \
283 SymI_HasProto(pow) \
284 SymI_HasProto(tanh) \
285 SymI_HasProto(cosh) \
286 SymI_HasProto(sinh) \
287 SymI_HasProto(atan) \
288 SymI_HasProto(acos) \
289 SymI_HasProto(asin) \
290 SymI_HasProto(tan) \
291 SymI_HasProto(cos) \
292 SymI_HasProto(sin) \
293 SymI_HasProto(exp) \
294 SymI_HasProto(log) \
295 SymI_HasProto(sqrt) \
296 SymI_HasProto(localtime_r) \
297 SymI_HasProto(gmtime_r) \
298 SymI_HasProto(mktime) \
299 SymI_NeedsProto(_imp___tzname) \
300 SymI_HasProto(gettimeofday) \
301 SymI_HasProto(timezone) \
302 SymI_HasProto(tcgetattr) \
303 SymI_HasProto(tcsetattr) \
304 SymI_HasProto(memcpy) \
305 SymI_HasProto(memmove) \
306 SymI_HasProto(realloc) \
307 SymI_HasProto(malloc) \
308 SymI_HasProto(free) \
309 SymI_HasProto(fork) \
310 SymI_HasProto(lstat) \
311 SymI_HasProto(isatty) \
312 SymI_HasProto(mkdir) \
313 SymI_HasProto(opendir) \
314 SymI_HasProto(readdir) \
315 SymI_HasProto(rewinddir) \
316 SymI_HasProto(closedir) \
317 SymI_HasProto(link) \
318 SymI_HasProto(mkfifo) \
319 SymI_HasProto(pipe) \
320 SymI_HasProto(read) \
321 SymI_HasProto(rename) \
322 SymI_HasProto(rmdir) \
323 SymI_HasProto(select) \
324 SymI_HasProto(system) \
325 SymI_HasProto(write) \
326 SymI_HasProto(strcmp) \
327 SymI_HasProto(strcpy) \
328 SymI_HasProto(strncpy) \
329 SymI_HasProto(strerror) \
330 SymI_HasProto(sigaddset) \
331 SymI_HasProto(sigemptyset) \
332 SymI_HasProto(sigprocmask) \
333 SymI_HasProto(umask) \
334 SymI_HasProto(uname) \
335 SymI_HasProto(unlink) \
336 SymI_HasProto(utime) \
337 SymI_HasProto(waitpid)
338
339 #elif !defined(mingw32_HOST_OS)
340 #define RTS_MINGW_ONLY_SYMBOLS /**/
341 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
342 #else /* defined(mingw32_HOST_OS) */
343 #define RTS_POSIX_ONLY_SYMBOLS /**/
344 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
345
346 /* Extra syms gen'ed by mingw-2's gcc-3.2: */
347 #if __GNUC__>=3
348 #define RTS_MINGW_EXTRA_SYMS \
349 SymI_NeedsProto(_imp____mb_cur_max) \
350 SymI_NeedsProto(_imp___pctype)
351 #else
352 #define RTS_MINGW_EXTRA_SYMS
353 #endif
354
355 #if HAVE_GETTIMEOFDAY
356 #define RTS_MINGW_GETTIMEOFDAY_SYM SymI_NeedsProto(gettimeofday)
357 #else
358 #define RTS_MINGW_GETTIMEOFDAY_SYM /**/
359 #endif
360
361 #if HAVE___MINGW_VFPRINTF
362 #define RTS___MINGW_VFPRINTF_SYM SymI_HasProto(__mingw_vfprintf)
363 #else
364 #define RTS___MINGW_VFPRINTF_SYM /**/
365 #endif
366
367 /* These are statically linked from the mingw libraries into the ghc
368 executable, so we have to employ this hack. */
369 #define RTS_MINGW_ONLY_SYMBOLS \
370 SymI_HasProto(stg_asyncReadzh) \
371 SymI_HasProto(stg_asyncWritezh) \
372 SymI_HasProto(stg_asyncDoProczh) \
373 SymI_HasProto(memset) \
374 SymI_HasProto(inet_ntoa) \
375 SymI_HasProto(inet_addr) \
376 SymI_HasProto(htonl) \
377 SymI_HasProto(recvfrom) \
378 SymI_HasProto(listen) \
379 SymI_HasProto(bind) \
380 SymI_HasProto(shutdown) \
381 SymI_HasProto(connect) \
382 SymI_HasProto(htons) \
383 SymI_HasProto(ntohs) \
384 SymI_HasProto(getservbyname) \
385 SymI_HasProto(getservbyport) \
386 SymI_HasProto(getprotobynumber) \
387 SymI_HasProto(getprotobyname) \
388 SymI_HasProto(gethostbyname) \
389 SymI_HasProto(gethostbyaddr) \
390 SymI_HasProto(gethostname) \
391 SymI_HasProto(strcpy) \
392 SymI_HasProto(strncpy) \
393 SymI_HasProto(abort) \
394 SymI_NeedsProto(_alloca) \
395 SymI_HasProto(isxdigit) \
396 SymI_HasProto(isupper) \
397 SymI_HasProto(ispunct) \
398 SymI_HasProto(islower) \
399 SymI_HasProto(isspace) \
400 SymI_HasProto(isprint) \
401 SymI_HasProto(isdigit) \
402 SymI_HasProto(iscntrl) \
403 SymI_HasProto(isalpha) \
404 SymI_HasProto(isalnum) \
405 SymI_HasProto(isascii) \
406 RTS___MINGW_VFPRINTF_SYM \
407 SymI_HasProto(strcmp) \
408 SymI_HasProto(memmove) \
409 SymI_HasProto(realloc) \
410 SymI_HasProto(malloc) \
411 SymI_HasProto(pow) \
412 SymI_HasProto(tanh) \
413 SymI_HasProto(cosh) \
414 SymI_HasProto(sinh) \
415 SymI_HasProto(atan) \
416 SymI_HasProto(acos) \
417 SymI_HasProto(asin) \
418 SymI_HasProto(tan) \
419 SymI_HasProto(cos) \
420 SymI_HasProto(sin) \
421 SymI_HasProto(exp) \
422 SymI_HasProto(log) \
423 SymI_HasProto(sqrt) \
424 SymI_HasProto(powf) \
425 SymI_HasProto(tanhf) \
426 SymI_HasProto(coshf) \
427 SymI_HasProto(sinhf) \
428 SymI_HasProto(atanf) \
429 SymI_HasProto(acosf) \
430 SymI_HasProto(asinf) \
431 SymI_HasProto(tanf) \
432 SymI_HasProto(cosf) \
433 SymI_HasProto(sinf) \
434 SymI_HasProto(expf) \
435 SymI_HasProto(logf) \
436 SymI_HasProto(sqrtf) \
437 SymI_HasProto(erf) \
438 SymI_HasProto(erfc) \
439 SymI_HasProto(erff) \
440 SymI_HasProto(erfcf) \
441 SymI_HasProto(memcpy) \
442 SymI_HasProto(rts_InstallConsoleEvent) \
443 SymI_HasProto(rts_ConsoleHandlerDone) \
444 SymI_NeedsProto(mktime) \
445 SymI_NeedsProto(_imp___timezone) \
446 SymI_NeedsProto(_imp___tzname) \
447 SymI_NeedsProto(_imp__tzname) \
448 SymI_NeedsProto(_imp___iob) \
449 SymI_NeedsProto(_imp___osver) \
450 SymI_NeedsProto(localtime) \
451 SymI_NeedsProto(gmtime) \
452 SymI_NeedsProto(opendir) \
453 SymI_NeedsProto(readdir) \
454 SymI_NeedsProto(rewinddir) \
455 RTS_MINGW_EXTRA_SYMS \
456 RTS_MINGW_GETTIMEOFDAY_SYM \
457 SymI_NeedsProto(closedir)
458 #endif
459
460 #if defined(darwin_HOST_OS) && HAVE_PRINTF_LDBLSTUB
461 #define RTS_DARWIN_ONLY_SYMBOLS \
462 SymI_NeedsProto(asprintf$LDBLStub) \
463 SymI_NeedsProto(err$LDBLStub) \
464 SymI_NeedsProto(errc$LDBLStub) \
465 SymI_NeedsProto(errx$LDBLStub) \
466 SymI_NeedsProto(fprintf$LDBLStub) \
467 SymI_NeedsProto(fscanf$LDBLStub) \
468 SymI_NeedsProto(fwprintf$LDBLStub) \
469 SymI_NeedsProto(fwscanf$LDBLStub) \
470 SymI_NeedsProto(printf$LDBLStub) \
471 SymI_NeedsProto(scanf$LDBLStub) \
472 SymI_NeedsProto(snprintf$LDBLStub) \
473 SymI_NeedsProto(sprintf$LDBLStub) \
474 SymI_NeedsProto(sscanf$LDBLStub) \
475 SymI_NeedsProto(strtold$LDBLStub) \
476 SymI_NeedsProto(swprintf$LDBLStub) \
477 SymI_NeedsProto(swscanf$LDBLStub) \
478 SymI_NeedsProto(syslog$LDBLStub) \
479 SymI_NeedsProto(vasprintf$LDBLStub) \
480 SymI_NeedsProto(verr$LDBLStub) \
481 SymI_NeedsProto(verrc$LDBLStub) \
482 SymI_NeedsProto(verrx$LDBLStub) \
483 SymI_NeedsProto(vfprintf$LDBLStub) \
484 SymI_NeedsProto(vfscanf$LDBLStub) \
485 SymI_NeedsProto(vfwprintf$LDBLStub) \
486 SymI_NeedsProto(vfwscanf$LDBLStub) \
487 SymI_NeedsProto(vprintf$LDBLStub) \
488 SymI_NeedsProto(vscanf$LDBLStub) \
489 SymI_NeedsProto(vsnprintf$LDBLStub) \
490 SymI_NeedsProto(vsprintf$LDBLStub) \
491 SymI_NeedsProto(vsscanf$LDBLStub) \
492 SymI_NeedsProto(vswprintf$LDBLStub) \
493 SymI_NeedsProto(vswscanf$LDBLStub) \
494 SymI_NeedsProto(vsyslog$LDBLStub) \
495 SymI_NeedsProto(vwarn$LDBLStub) \
496 SymI_NeedsProto(vwarnc$LDBLStub) \
497 SymI_NeedsProto(vwarnx$LDBLStub) \
498 SymI_NeedsProto(vwprintf$LDBLStub) \
499 SymI_NeedsProto(vwscanf$LDBLStub) \
500 SymI_NeedsProto(warn$LDBLStub) \
501 SymI_NeedsProto(warnc$LDBLStub) \
502 SymI_NeedsProto(warnx$LDBLStub) \
503 SymI_NeedsProto(wcstold$LDBLStub) \
504 SymI_NeedsProto(wprintf$LDBLStub) \
505 SymI_NeedsProto(wscanf$LDBLStub)
506 #else
507 #define RTS_DARWIN_ONLY_SYMBOLS
508 #endif
509
510 #ifndef SMP
511 # define MAIN_CAP_SYM SymI_HasProto(MainCapability)
512 #else
513 # define MAIN_CAP_SYM
514 #endif
515
516 #if !defined(mingw32_HOST_OS)
517 #define RTS_USER_SIGNALS_SYMBOLS \
518 SymI_HasProto(setIOManagerPipe) \
519 SymI_HasProto(ioManagerWakeup) \
520 SymI_HasProto(ioManagerSync) \
521 SymI_HasProto(blockUserSignals) \
522 SymI_HasProto(unblockUserSignals)
523 #else
524 #define RTS_USER_SIGNALS_SYMBOLS \
525 SymI_HasProto(ioManagerWakeup) \
526 SymI_HasProto(sendIOManagerEvent) \
527 SymI_HasProto(readIOManagerEvent) \
528 SymI_HasProto(getIOManagerEvent) \
529 SymI_HasProto(console_handler)
530 #endif
531
532 #define RTS_LIBFFI_SYMBOLS \
533 SymE_NeedsProto(ffi_prep_cif) \
534 SymE_NeedsProto(ffi_call) \
535 SymE_NeedsProto(ffi_type_void) \
536 SymE_NeedsProto(ffi_type_float) \
537 SymE_NeedsProto(ffi_type_double) \
538 SymE_NeedsProto(ffi_type_sint64) \
539 SymE_NeedsProto(ffi_type_uint64) \
540 SymE_NeedsProto(ffi_type_sint32) \
541 SymE_NeedsProto(ffi_type_uint32) \
542 SymE_NeedsProto(ffi_type_sint16) \
543 SymE_NeedsProto(ffi_type_uint16) \
544 SymE_NeedsProto(ffi_type_sint8) \
545 SymE_NeedsProto(ffi_type_uint8) \
546 SymE_NeedsProto(ffi_type_pointer)
547
548 #ifdef TABLES_NEXT_TO_CODE
549 #define RTS_RET_SYMBOLS /* nothing */
550 #else
551 #define RTS_RET_SYMBOLS \
552 SymI_HasProto(stg_enter_ret) \
553 SymI_HasProto(stg_gc_fun_ret) \
554 SymI_HasProto(stg_ap_v_ret) \
555 SymI_HasProto(stg_ap_f_ret) \
556 SymI_HasProto(stg_ap_d_ret) \
557 SymI_HasProto(stg_ap_l_ret) \
558 SymI_HasProto(stg_ap_n_ret) \
559 SymI_HasProto(stg_ap_p_ret) \
560 SymI_HasProto(stg_ap_pv_ret) \
561 SymI_HasProto(stg_ap_pp_ret) \
562 SymI_HasProto(stg_ap_ppv_ret) \
563 SymI_HasProto(stg_ap_ppp_ret) \
564 SymI_HasProto(stg_ap_pppv_ret) \
565 SymI_HasProto(stg_ap_pppp_ret) \
566 SymI_HasProto(stg_ap_ppppp_ret) \
567 SymI_HasProto(stg_ap_pppppp_ret)
568 #endif
569
570 /* Modules compiled with -ticky may mention ticky counters */
571 /* This list should marry up with the one in $(TOP)/includes/stg/Ticky.h */
572 #define RTS_TICKY_SYMBOLS \
573 SymI_NeedsProto(ticky_entry_ctrs) \
574 SymI_NeedsProto(top_ct) \
575 \
576 SymI_HasProto(ENT_VIA_NODE_ctr) \
577 SymI_HasProto(ENT_STATIC_THK_ctr) \
578 SymI_HasProto(ENT_DYN_THK_ctr) \
579 SymI_HasProto(ENT_STATIC_FUN_DIRECT_ctr) \
580 SymI_HasProto(ENT_DYN_FUN_DIRECT_ctr) \
581 SymI_HasProto(ENT_STATIC_CON_ctr) \
582 SymI_HasProto(ENT_DYN_CON_ctr) \
583 SymI_HasProto(ENT_STATIC_IND_ctr) \
584 SymI_HasProto(ENT_DYN_IND_ctr) \
585 SymI_HasProto(ENT_PERM_IND_ctr) \
586 SymI_HasProto(ENT_PAP_ctr) \
587 SymI_HasProto(ENT_AP_ctr) \
588 SymI_HasProto(ENT_AP_STACK_ctr) \
589 SymI_HasProto(ENT_BH_ctr) \
590 SymI_HasProto(UNKNOWN_CALL_ctr) \
591 SymI_HasProto(SLOW_CALL_v_ctr) \
592 SymI_HasProto(SLOW_CALL_f_ctr) \
593 SymI_HasProto(SLOW_CALL_d_ctr) \
594 SymI_HasProto(SLOW_CALL_l_ctr) \
595 SymI_HasProto(SLOW_CALL_n_ctr) \
596 SymI_HasProto(SLOW_CALL_p_ctr) \
597 SymI_HasProto(SLOW_CALL_pv_ctr) \
598 SymI_HasProto(SLOW_CALL_pp_ctr) \
599 SymI_HasProto(SLOW_CALL_ppv_ctr) \
600 SymI_HasProto(SLOW_CALL_ppp_ctr) \
601 SymI_HasProto(SLOW_CALL_pppv_ctr) \
602 SymI_HasProto(SLOW_CALL_pppp_ctr) \
603 SymI_HasProto(SLOW_CALL_ppppp_ctr) \
604 SymI_HasProto(SLOW_CALL_pppppp_ctr) \
605 SymI_HasProto(SLOW_CALL_OTHER_ctr) \
606 SymI_HasProto(ticky_slow_call_unevald) \
607 SymI_HasProto(SLOW_CALL_ctr) \
608 SymI_HasProto(MULTI_CHUNK_SLOW_CALL_ctr) \
609 SymI_HasProto(MULTI_CHUNK_SLOW_CALL_CHUNKS_ctr) \
610 SymI_HasProto(KNOWN_CALL_ctr) \
611 SymI_HasProto(KNOWN_CALL_TOO_FEW_ARGS_ctr) \
612 SymI_HasProto(KNOWN_CALL_EXTRA_ARGS_ctr) \
613 SymI_HasProto(SLOW_CALL_FUN_TOO_FEW_ctr) \
614 SymI_HasProto(SLOW_CALL_FUN_CORRECT_ctr) \
615 SymI_HasProto(SLOW_CALL_FUN_TOO_MANY_ctr) \
616 SymI_HasProto(SLOW_CALL_PAP_TOO_FEW_ctr) \
617 SymI_HasProto(SLOW_CALL_PAP_CORRECT_ctr) \
618 SymI_HasProto(SLOW_CALL_PAP_TOO_MANY_ctr) \
619 SymI_HasProto(SLOW_CALL_UNEVALD_ctr) \
620 SymI_HasProto(UPDF_OMITTED_ctr) \
621 SymI_HasProto(UPDF_PUSHED_ctr) \
622 SymI_HasProto(CATCHF_PUSHED_ctr) \
623 SymI_HasProto(UPDF_RCC_PUSHED_ctr) \
624 SymI_HasProto(UPDF_RCC_OMITTED_ctr) \
625 SymI_HasProto(UPD_SQUEEZED_ctr) \
626 SymI_HasProto(UPD_CON_IN_NEW_ctr) \
627 SymI_HasProto(UPD_CON_IN_PLACE_ctr) \
628 SymI_HasProto(UPD_PAP_IN_NEW_ctr) \
629 SymI_HasProto(UPD_PAP_IN_PLACE_ctr) \
630 SymI_HasProto(ALLOC_HEAP_ctr) \
631 SymI_HasProto(ALLOC_HEAP_tot) \
632 SymI_HasProto(ALLOC_FUN_ctr) \
633 SymI_HasProto(ALLOC_FUN_adm) \
634 SymI_HasProto(ALLOC_FUN_gds) \
635 SymI_HasProto(ALLOC_FUN_slp) \
636 SymI_HasProto(UPD_NEW_IND_ctr) \
637 SymI_HasProto(UPD_NEW_PERM_IND_ctr) \
638 SymI_HasProto(UPD_OLD_IND_ctr) \
639 SymI_HasProto(UPD_OLD_PERM_IND_ctr) \
640 SymI_HasProto(UPD_BH_UPDATABLE_ctr) \
641 SymI_HasProto(UPD_BH_SINGLE_ENTRY_ctr) \
642 SymI_HasProto(UPD_CAF_BH_UPDATABLE_ctr) \
643 SymI_HasProto(UPD_CAF_BH_SINGLE_ENTRY_ctr) \
644 SymI_HasProto(GC_SEL_ABANDONED_ctr) \
645 SymI_HasProto(GC_SEL_MINOR_ctr) \
646 SymI_HasProto(GC_SEL_MAJOR_ctr) \
647 SymI_HasProto(GC_FAILED_PROMOTION_ctr) \
648 SymI_HasProto(ALLOC_UP_THK_ctr) \
649 SymI_HasProto(ALLOC_SE_THK_ctr) \
650 SymI_HasProto(ALLOC_THK_adm) \
651 SymI_HasProto(ALLOC_THK_gds) \
652 SymI_HasProto(ALLOC_THK_slp) \
653 SymI_HasProto(ALLOC_CON_ctr) \
654 SymI_HasProto(ALLOC_CON_adm) \
655 SymI_HasProto(ALLOC_CON_gds) \
656 SymI_HasProto(ALLOC_CON_slp) \
657 SymI_HasProto(ALLOC_TUP_ctr) \
658 SymI_HasProto(ALLOC_TUP_adm) \
659 SymI_HasProto(ALLOC_TUP_gds) \
660 SymI_HasProto(ALLOC_TUP_slp) \
661 SymI_HasProto(ALLOC_BH_ctr) \
662 SymI_HasProto(ALLOC_BH_adm) \
663 SymI_HasProto(ALLOC_BH_gds) \
664 SymI_HasProto(ALLOC_BH_slp) \
665 SymI_HasProto(ALLOC_PRIM_ctr) \
666 SymI_HasProto(ALLOC_PRIM_adm) \
667 SymI_HasProto(ALLOC_PRIM_gds) \
668 SymI_HasProto(ALLOC_PRIM_slp) \
669 SymI_HasProto(ALLOC_PAP_ctr) \
670 SymI_HasProto(ALLOC_PAP_adm) \
671 SymI_HasProto(ALLOC_PAP_gds) \
672 SymI_HasProto(ALLOC_PAP_slp) \
673 SymI_HasProto(ALLOC_TSO_ctr) \
674 SymI_HasProto(ALLOC_TSO_adm) \
675 SymI_HasProto(ALLOC_TSO_gds) \
676 SymI_HasProto(ALLOC_TSO_slp) \
677 SymI_HasProto(RET_NEW_ctr) \
678 SymI_HasProto(RET_OLD_ctr) \
679 SymI_HasProto(RET_UNBOXED_TUP_ctr) \
680 SymI_HasProto(RET_SEMI_loads_avoided)
681
682
683 // On most platforms, the garbage collector rewrites references
684 // to small integer and char objects to a set of common, shared ones.
685 //
686 // We don't do this when compiling to Windows DLLs at the moment because
687 // it doesn't support cross package data references well.
688 //
689 #if defined(__PIC__) && defined(mingw32_HOST_OS)
690 #define RTS_INTCHAR_SYMBOLS
691 #else
692 #define RTS_INTCHAR_SYMBOLS \
693 SymI_HasProto(stg_CHARLIKE_closure) \
694 SymI_HasProto(stg_INTLIKE_closure)
695 #endif
696
697
698 #define RTS_SYMBOLS \
699 Maybe_Stable_Names \
700 RTS_TICKY_SYMBOLS \
701 SymI_HasProto(StgReturn) \
702 SymI_HasProto(stg_enter_info) \
703 SymI_HasProto(stg_gc_void_info) \
704 SymI_HasProto(__stg_gc_enter_1) \
705 SymI_HasProto(stg_gc_noregs) \
706 SymI_HasProto(stg_gc_unpt_r1_info) \
707 SymI_HasProto(stg_gc_unpt_r1) \
708 SymI_HasProto(stg_gc_unbx_r1_info) \
709 SymI_HasProto(stg_gc_unbx_r1) \
710 SymI_HasProto(stg_gc_f1_info) \
711 SymI_HasProto(stg_gc_f1) \
712 SymI_HasProto(stg_gc_d1_info) \
713 SymI_HasProto(stg_gc_d1) \
714 SymI_HasProto(stg_gc_l1_info) \
715 SymI_HasProto(stg_gc_l1) \
716 SymI_HasProto(__stg_gc_fun) \
717 SymI_HasProto(stg_gc_fun_info) \
718 SymI_HasProto(stg_gc_gen) \
719 SymI_HasProto(stg_gc_gen_info) \
720 SymI_HasProto(stg_gc_gen_hp) \
721 SymI_HasProto(stg_gc_ut) \
722 SymI_HasProto(stg_gen_yield) \
723 SymI_HasProto(stg_yield_noregs) \
724 SymI_HasProto(stg_yield_to_interpreter) \
725 SymI_HasProto(stg_gen_block) \
726 SymI_HasProto(stg_block_noregs) \
727 SymI_HasProto(stg_block_1) \
728 SymI_HasProto(stg_block_takemvar) \
729 SymI_HasProto(stg_block_putmvar) \
730 MAIN_CAP_SYM \
731 SymI_HasProto(MallocFailHook) \
732 SymI_HasProto(OnExitHook) \
733 SymI_HasProto(OutOfHeapHook) \
734 SymI_HasProto(StackOverflowHook) \
735 SymI_HasProto(addDLL) \
736 SymI_HasProto(__int_encodeDouble) \
737 SymI_HasProto(__word_encodeDouble) \
738 SymI_HasProto(__2Int_encodeDouble) \
739 SymI_HasProto(__int_encodeFloat) \
740 SymI_HasProto(__word_encodeFloat) \
741 SymI_HasProto(stg_atomicallyzh) \
742 SymI_HasProto(barf) \
743 SymI_HasProto(debugBelch) \
744 SymI_HasProto(errorBelch) \
745 SymI_HasProto(sysErrorBelch) \
746 SymI_HasProto(stg_getMaskingStatezh) \
747 SymI_HasProto(stg_maskAsyncExceptionszh) \
748 SymI_HasProto(stg_maskUninterruptiblezh) \
749 SymI_HasProto(stg_catchzh) \
750 SymI_HasProto(stg_catchRetryzh) \
751 SymI_HasProto(stg_catchSTMzh) \
752 SymI_HasProto(stg_checkzh) \
753 SymI_HasProto(closure_flags) \
754 SymI_HasProto(cmp_thread) \
755 SymI_HasProto(createAdjustor) \
756 SymI_HasProto(stg_decodeDoublezu2Intzh) \
757 SymI_HasProto(stg_decodeFloatzuIntzh) \
758 SymI_HasProto(defaultsHook) \
759 SymI_HasProto(stg_delayzh) \
760 SymI_HasProto(stg_deRefWeakzh) \
761 SymI_HasProto(stg_deRefStablePtrzh) \
762 SymI_HasProto(dirty_MUT_VAR) \
763 SymI_HasProto(stg_forkzh) \
764 SymI_HasProto(stg_forkOnzh) \
765 SymI_HasProto(forkProcess) \
766 SymI_HasProto(forkOS_createThread) \
767 SymI_HasProto(freeHaskellFunctionPtr) \
768 SymI_HasProto(getOrSetTypeableStore) \
769 SymI_HasProto(getOrSetGHCConcSignalHandlerStore) \
770 SymI_HasProto(getOrSetGHCConcPendingEventsStore) \
771 SymI_HasProto(getOrSetGHCConcPendingDelaysStore) \
772 SymI_HasProto(getOrSetGHCConcIOManagerThreadStore) \
773 SymI_HasProto(getOrSetGHCConcProddingStore) \
774 SymI_HasProto(genSymZh) \
775 SymI_HasProto(genericRaise) \
776 SymI_HasProto(getProgArgv) \
777 SymI_HasProto(getFullProgArgv) \
778 SymI_HasProto(getStablePtr) \
779 SymI_HasProto(hs_init) \
780 SymI_HasProto(hs_exit) \
781 SymI_HasProto(hs_set_argv) \
782 SymI_HasProto(hs_add_root) \
783 SymI_HasProto(hs_perform_gc) \
784 SymI_HasProto(hs_free_stable_ptr) \
785 SymI_HasProto(hs_free_fun_ptr) \
786 SymI_HasProto(hs_hpc_rootModule) \
787 SymI_HasProto(hs_hpc_module) \
788 SymI_HasProto(initLinker) \
789 SymI_HasProto(stg_unpackClosurezh) \
790 SymI_HasProto(stg_getApStackValzh) \
791 SymI_HasProto(stg_getSparkzh) \
792 SymI_HasProto(stg_numSparkszh) \
793 SymI_HasProto(stg_isCurrentThreadBoundzh) \
794 SymI_HasProto(stg_isEmptyMVarzh) \
795 SymI_HasProto(stg_killThreadzh) \
796 SymI_HasProto(loadObj) \
797 SymI_HasProto(insertStableSymbol) \
798 SymI_HasProto(insertSymbol) \
799 SymI_HasProto(lookupSymbol) \
800 SymI_HasProto(stg_makeStablePtrzh) \
801 SymI_HasProto(stg_mkApUpd0zh) \
802 SymI_HasProto(stg_myThreadIdzh) \
803 SymI_HasProto(stg_labelThreadzh) \
804 SymI_HasProto(stg_newArrayzh) \
805 SymI_HasProto(stg_newBCOzh) \
806 SymI_HasProto(stg_newByteArrayzh) \
807 SymI_HasProto_redirect(newCAF, newDynCAF) \
808 SymI_HasProto(stg_newMVarzh) \
809 SymI_HasProto(stg_newMutVarzh) \
810 SymI_HasProto(stg_newTVarzh) \
811 SymI_HasProto(stg_noDuplicatezh) \
812 SymI_HasProto(stg_atomicModifyMutVarzh) \
813 SymI_HasProto(stg_newPinnedByteArrayzh) \
814 SymI_HasProto(stg_newAlignedPinnedByteArrayzh) \
815 SymI_HasProto(newSpark) \
816 SymI_HasProto(performGC) \
817 SymI_HasProto(performMajorGC) \
818 SymI_HasProto(prog_argc) \
819 SymI_HasProto(prog_argv) \
820 SymI_HasProto(stg_putMVarzh) \
821 SymI_HasProto(stg_raisezh) \
822 SymI_HasProto(stg_raiseIOzh) \
823 SymI_HasProto(stg_readTVarzh) \
824 SymI_HasProto(stg_readTVarIOzh) \
825 SymI_HasProto(resumeThread) \
826 SymI_HasProto(resolveObjs) \
827 SymI_HasProto(stg_retryzh) \
828 SymI_HasProto(rts_apply) \
829 SymI_HasProto(rts_checkSchedStatus) \
830 SymI_HasProto(rts_eval) \
831 SymI_HasProto(rts_evalIO) \
832 SymI_HasProto(rts_evalLazyIO) \
833 SymI_HasProto(rts_evalStableIO) \
834 SymI_HasProto(rts_eval_) \
835 SymI_HasProto(rts_getBool) \
836 SymI_HasProto(rts_getChar) \
837 SymI_HasProto(rts_getDouble) \
838 SymI_HasProto(rts_getFloat) \
839 SymI_HasProto(rts_getInt) \
840 SymI_HasProto(rts_getInt8) \
841 SymI_HasProto(rts_getInt16) \
842 SymI_HasProto(rts_getInt32) \
843 SymI_HasProto(rts_getInt64) \
844 SymI_HasProto(rts_getPtr) \
845 SymI_HasProto(rts_getFunPtr) \
846 SymI_HasProto(rts_getStablePtr) \
847 SymI_HasProto(rts_getThreadId) \
848 SymI_HasProto(rts_getWord) \
849 SymI_HasProto(rts_getWord8) \
850 SymI_HasProto(rts_getWord16) \
851 SymI_HasProto(rts_getWord32) \
852 SymI_HasProto(rts_getWord64) \
853 SymI_HasProto(rts_lock) \
854 SymI_HasProto(rts_mkBool) \
855 SymI_HasProto(rts_mkChar) \
856 SymI_HasProto(rts_mkDouble) \
857 SymI_HasProto(rts_mkFloat) \
858 SymI_HasProto(rts_mkInt) \
859 SymI_HasProto(rts_mkInt8) \
860 SymI_HasProto(rts_mkInt16) \
861 SymI_HasProto(rts_mkInt32) \
862 SymI_HasProto(rts_mkInt64) \
863 SymI_HasProto(rts_mkPtr) \
864 SymI_HasProto(rts_mkFunPtr) \
865 SymI_HasProto(rts_mkStablePtr) \
866 SymI_HasProto(rts_mkString) \
867 SymI_HasProto(rts_mkWord) \
868 SymI_HasProto(rts_mkWord8) \
869 SymI_HasProto(rts_mkWord16) \
870 SymI_HasProto(rts_mkWord32) \
871 SymI_HasProto(rts_mkWord64) \
872 SymI_HasProto(rts_unlock) \
873 SymI_HasProto(rts_unsafeGetMyCapability) \
874 SymI_HasProto(rtsSupportsBoundThreads) \
875 SymI_HasProto(setProgArgv) \
876 SymI_HasProto(startupHaskell) \
877 SymI_HasProto(shutdownHaskell) \
878 SymI_HasProto(shutdownHaskellAndExit) \
879 SymI_HasProto(stable_ptr_table) \
880 SymI_HasProto(stackOverflow) \
881 SymI_HasProto(stg_CAF_BLACKHOLE_info) \
882 SymI_HasProto(stg_BLACKHOLE_info) \
883 SymI_HasProto(__stg_EAGER_BLACKHOLE_info) \
884 SymI_HasProto(stg_BLOCKING_QUEUE_CLEAN_info) \
885 SymI_HasProto(stg_BLOCKING_QUEUE_DIRTY_info) \
886 SymI_HasProto(startTimer) \
887 SymI_HasProto(stg_MVAR_CLEAN_info) \
888 SymI_HasProto(stg_MVAR_DIRTY_info) \
889 SymI_HasProto(stg_IND_STATIC_info) \
890 SymI_HasProto(stg_ARR_WORDS_info) \
891 SymI_HasProto(stg_MUT_ARR_PTRS_DIRTY_info) \
892 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN_info) \
893 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN0_info) \
894 SymI_HasProto(stg_WEAK_info) \
895 SymI_HasProto(stg_ap_v_info) \
896 SymI_HasProto(stg_ap_f_info) \
897 SymI_HasProto(stg_ap_d_info) \
898 SymI_HasProto(stg_ap_l_info) \
899 SymI_HasProto(stg_ap_n_info) \
900 SymI_HasProto(stg_ap_p_info) \
901 SymI_HasProto(stg_ap_pv_info) \
902 SymI_HasProto(stg_ap_pp_info) \
903 SymI_HasProto(stg_ap_ppv_info) \
904 SymI_HasProto(stg_ap_ppp_info) \
905 SymI_HasProto(stg_ap_pppv_info) \
906 SymI_HasProto(stg_ap_pppp_info) \
907 SymI_HasProto(stg_ap_ppppp_info) \
908 SymI_HasProto(stg_ap_pppppp_info) \
909 SymI_HasProto(stg_ap_0_fast) \
910 SymI_HasProto(stg_ap_v_fast) \
911 SymI_HasProto(stg_ap_f_fast) \
912 SymI_HasProto(stg_ap_d_fast) \
913 SymI_HasProto(stg_ap_l_fast) \
914 SymI_HasProto(stg_ap_n_fast) \
915 SymI_HasProto(stg_ap_p_fast) \
916 SymI_HasProto(stg_ap_pv_fast) \
917 SymI_HasProto(stg_ap_pp_fast) \
918 SymI_HasProto(stg_ap_ppv_fast) \
919 SymI_HasProto(stg_ap_ppp_fast) \
920 SymI_HasProto(stg_ap_pppv_fast) \
921 SymI_HasProto(stg_ap_pppp_fast) \
922 SymI_HasProto(stg_ap_ppppp_fast) \
923 SymI_HasProto(stg_ap_pppppp_fast) \
924 SymI_HasProto(stg_ap_1_upd_info) \
925 SymI_HasProto(stg_ap_2_upd_info) \
926 SymI_HasProto(stg_ap_3_upd_info) \
927 SymI_HasProto(stg_ap_4_upd_info) \
928 SymI_HasProto(stg_ap_5_upd_info) \
929 SymI_HasProto(stg_ap_6_upd_info) \
930 SymI_HasProto(stg_ap_7_upd_info) \
931 SymI_HasProto(stg_exit) \
932 SymI_HasProto(stg_sel_0_upd_info) \
933 SymI_HasProto(stg_sel_10_upd_info) \
934 SymI_HasProto(stg_sel_11_upd_info) \
935 SymI_HasProto(stg_sel_12_upd_info) \
936 SymI_HasProto(stg_sel_13_upd_info) \
937 SymI_HasProto(stg_sel_14_upd_info) \
938 SymI_HasProto(stg_sel_15_upd_info) \
939 SymI_HasProto(stg_sel_1_upd_info) \
940 SymI_HasProto(stg_sel_2_upd_info) \
941 SymI_HasProto(stg_sel_3_upd_info) \
942 SymI_HasProto(stg_sel_4_upd_info) \
943 SymI_HasProto(stg_sel_5_upd_info) \
944 SymI_HasProto(stg_sel_6_upd_info) \
945 SymI_HasProto(stg_sel_7_upd_info) \
946 SymI_HasProto(stg_sel_8_upd_info) \
947 SymI_HasProto(stg_sel_9_upd_info) \
948 SymI_HasProto(stg_upd_frame_info) \
949 SymI_HasProto(stg_bh_upd_frame_info) \
950 SymI_HasProto(suspendThread) \
951 SymI_HasProto(stg_takeMVarzh) \
952 SymI_HasProto(stg_threadStatuszh) \
953 SymI_HasProto(stg_tryPutMVarzh) \
954 SymI_HasProto(stg_tryTakeMVarzh) \
955 SymI_HasProto(stg_unmaskAsyncExceptionszh) \
956 SymI_HasProto(unloadObj) \
957 SymI_HasProto(stg_unsafeThawArrayzh) \
958 SymI_HasProto(stg_waitReadzh) \
959 SymI_HasProto(stg_waitWritezh) \
960 SymI_HasProto(stg_writeTVarzh) \
961 SymI_HasProto(stg_yieldzh) \
962 SymI_NeedsProto(stg_interp_constr_entry) \
963 SymI_HasProto(alloc_blocks_lim) \
964 SymI_HasProto(g0) \
965 SymI_HasProto(allocate) \
966 SymI_HasProto(allocateExec) \
967 SymI_HasProto(freeExec) \
968 SymI_HasProto(getAllocations) \
969 SymI_HasProto(revertCAFs) \
970 SymI_HasProto(RtsFlags) \
971 SymI_NeedsProto(rts_breakpoint_io_action) \
972 SymI_NeedsProto(rts_stop_next_breakpoint) \
973 SymI_NeedsProto(rts_stop_on_exception) \
974 SymI_HasProto(stopTimer) \
975 SymI_HasProto(n_capabilities) \
976 SymI_HasProto(stg_traceCcszh) \
977 SymI_HasProto(stg_traceEventzh) \
978 RTS_USER_SIGNALS_SYMBOLS \
979 RTS_INTCHAR_SYMBOLS
980
981
982 // 64-bit support functions in libgcc.a
983 #if defined(__GNUC__) && SIZEOF_VOID_P <= 4
984 #define RTS_LIBGCC_SYMBOLS \
985 SymI_NeedsProto(__divdi3) \
986 SymI_NeedsProto(__udivdi3) \
987 SymI_NeedsProto(__moddi3) \
988 SymI_NeedsProto(__umoddi3) \
989 SymI_NeedsProto(__muldi3) \
990 SymI_NeedsProto(__ashldi3) \
991 SymI_NeedsProto(__ashrdi3) \
992 SymI_NeedsProto(__lshrdi3)
993 #else
994 #define RTS_LIBGCC_SYMBOLS
995 #endif
996
997 #if defined(darwin_HOST_OS) && defined(powerpc_HOST_ARCH)
998 // Symbols that don't have a leading underscore
999 // on Mac OS X. They have to receive special treatment,
1000 // see machoInitSymbolsWithoutUnderscore()
1001 #define RTS_MACHO_NOUNDERLINE_SYMBOLS \
1002 SymI_NeedsProto(saveFP) \
1003 SymI_NeedsProto(restFP)
1004 #endif
1005
1006 /* entirely bogus claims about types of these symbols */
1007 #define SymI_NeedsProto(vvv) extern void vvv(void);
1008 #if defined(__PIC__) && defined(mingw32_HOST_OS)
1009 #define SymE_HasProto(vvv) SymE_HasProto(vvv);
1010 #define SymE_NeedsProto(vvv) extern void _imp__ ## vvv (void);
1011 #else
1012 #define SymE_NeedsProto(vvv) SymI_NeedsProto(vvv);
1013 #define SymE_HasProto(vvv) SymI_HasProto(vvv)
1014 #endif
1015 #define SymI_HasProto(vvv) /**/
1016 #define SymI_HasProto_redirect(vvv,xxx) /**/
1017 RTS_SYMBOLS
1018 RTS_RET_SYMBOLS
1019 RTS_POSIX_ONLY_SYMBOLS
1020 RTS_MINGW_ONLY_SYMBOLS
1021 RTS_CYGWIN_ONLY_SYMBOLS
1022 RTS_DARWIN_ONLY_SYMBOLS
1023 RTS_LIBGCC_SYMBOLS
1024 RTS_LIBFFI_SYMBOLS
1025 #undef SymI_NeedsProto
1026 #undef SymI_HasProto
1027 #undef SymI_HasProto_redirect
1028 #undef SymE_HasProto
1029 #undef SymE_NeedsProto
1030
1031 #ifdef LEADING_UNDERSCORE
1032 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
1033 #else
1034 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
1035 #endif
1036
1037 #define SymI_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1038 (void*)(&(vvv)) },
1039 #define SymE_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1040 (void*)DLL_IMPORT_DATA_REF(vvv) },
1041
1042 #define SymI_NeedsProto(vvv) SymI_HasProto(vvv)
1043 #define SymE_NeedsProto(vvv) SymE_HasProto(vvv)
1044
1045 // SymI_HasProto_redirect allows us to redirect references to one symbol to
1046 // another symbol. See newCAF/newDynCAF for an example.
1047 #define SymI_HasProto_redirect(vvv,xxx) \
1048 { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1049 (void*)(&(xxx)) },
1050
1051 static RtsSymbolVal rtsSyms[] = {
1052 RTS_SYMBOLS
1053 RTS_RET_SYMBOLS
1054 RTS_POSIX_ONLY_SYMBOLS
1055 RTS_MINGW_ONLY_SYMBOLS
1056 RTS_CYGWIN_ONLY_SYMBOLS
1057 RTS_DARWIN_ONLY_SYMBOLS
1058 RTS_LIBGCC_SYMBOLS
1059 RTS_LIBFFI_SYMBOLS
1060 #if defined(darwin_HOST_OS) && defined(i386_HOST_ARCH)
1061 // dyld stub code contains references to this,
1062 // but it should never be called because we treat
1063 // lazy pointers as nonlazy.
1064 { "dyld_stub_binding_helper", (void*)0xDEADBEEF },
1065 #endif
1066 { 0, 0 } /* sentinel */
1067 };
1068
1069
1070
1071 /* -----------------------------------------------------------------------------
1072 * Insert symbols into hash tables, checking for duplicates.
1073 */
1074
1075 static void ghciInsertStrHashTable ( char* obj_name,
1076 HashTable *table,
1077 char* key,
1078 void *data
1079 )
1080 {
1081 if (lookupHashTable(table, (StgWord)key) == NULL)
1082 {
1083 insertStrHashTable(table, (StgWord)key, data);
1084 return;
1085 }
1086 debugBelch(
1087 "\n\n"
1088 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
1089 " %s\n"
1090 "whilst processing object file\n"
1091 " %s\n"
1092 "This could be caused by:\n"
1093 " * Loading two different object files which export the same symbol\n"
1094 " * Specifying the same object file twice on the GHCi command line\n"
1095 " * An incorrect `package.conf' entry, causing some object to be\n"
1096 " loaded twice.\n"
1097 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
1098 "\n",
1099 (char*)key,
1100 obj_name
1101 );
1102 exit(1);
1103 }
1104 /* -----------------------------------------------------------------------------
1105 * initialize the object linker
1106 */
1107
1108
1109 static int linker_init_done = 0 ;
1110
1111 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1112 static void *dl_prog_handle;
1113 static regex_t re_invalid;
1114 static regex_t re_realso;
1115 #ifdef THREADED_RTS
1116 static Mutex dl_mutex; // mutex to protect dlopen/dlerror critical section
1117 #endif
1118 #endif
1119
1120 void
1121 initLinker( void )
1122 {
1123 RtsSymbolVal *sym;
1124 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1125 int compileResult;
1126 #endif
1127
1128 /* Make initLinker idempotent, so we can call it
1129 before evey relevant operation; that means we
1130 don't need to initialise the linker separately */
1131 if (linker_init_done == 1) { return; } else {
1132 linker_init_done = 1;
1133 }
1134
1135 #if defined(THREADED_RTS) && (defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO))
1136 initMutex(&dl_mutex);
1137 #endif
1138 stablehash = allocStrHashTable();
1139 symhash = allocStrHashTable();
1140
1141 /* populate the symbol table with stuff from the RTS */
1142 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
1143 ghciInsertStrHashTable("(GHCi built-in symbols)",
1144 symhash, sym->lbl, sym->addr);
1145 }
1146 # if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH)
1147 machoInitSymbolsWithoutUnderscore();
1148 # endif
1149
1150 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1151 # if defined(RTLD_DEFAULT)
1152 dl_prog_handle = RTLD_DEFAULT;
1153 # else
1154 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
1155 # endif /* RTLD_DEFAULT */
1156
1157 compileResult = regcomp(&re_invalid,
1158 "(([^ \t()])+\\.so([^ \t:()])*):([ \t])*invalid ELF header",
1159 REG_EXTENDED);
1160 ASSERT( compileResult == 0 );
1161 compileResult = regcomp(&re_realso,
1162 "GROUP *\\( *(([^ )])+)",
1163 REG_EXTENDED);
1164 ASSERT( compileResult == 0 );
1165 # endif
1166
1167 #if defined(x86_64_HOST_ARCH)
1168 if (RtsFlags.MiscFlags.linkerMemBase != 0) {
1169 // User-override for mmap_32bit_base
1170 mmap_32bit_base = (void*)RtsFlags.MiscFlags.linkerMemBase;
1171 }
1172 #endif
1173
1174 #if defined(mingw32_HOST_OS)
1175 /*
1176 * These two libraries cause problems when added to the static link,
1177 * but are necessary for resolving symbols in GHCi, hence we load
1178 * them manually here.
1179 */
1180 addDLL("msvcrt");
1181 addDLL("kernel32");
1182 #endif
1183 }
1184
1185 void
1186 exitLinker( void ) {
1187 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1188 if (linker_init_done == 1) {
1189 regfree(&re_invalid);
1190 regfree(&re_realso);
1191 #ifdef THREADED_RTS
1192 closeMutex(&dl_mutex);
1193 #endif
1194 }
1195 #endif
1196 }
1197
1198 /* -----------------------------------------------------------------------------
1199 * Loading DLL or .so dynamic libraries
1200 * -----------------------------------------------------------------------------
1201 *
1202 * Add a DLL from which symbols may be found. In the ELF case, just
1203 * do RTLD_GLOBAL-style add, so no further messing around needs to
1204 * happen in order that symbols in the loaded .so are findable --
1205 * lookupSymbol() will subsequently see them by dlsym on the program's
1206 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
1207 *
1208 * In the PEi386 case, open the DLLs and put handles to them in a
1209 * linked list. When looking for a symbol, try all handles in the
1210 * list. This means that we need to load even DLLs that are guaranteed
1211 * to be in the ghc.exe image already, just so we can get a handle
1212 * to give to loadSymbol, so that we can find the symbols. For such
1213 * libraries, the LoadLibrary call should be a no-op except for returning
1214 * the handle.
1215 *
1216 */
1217
1218 #if defined(OBJFORMAT_PEi386)
1219 /* A record for storing handles into DLLs. */
1220
1221 typedef
1222 struct _OpenedDLL {
1223 char* name;
1224 struct _OpenedDLL* next;
1225 HINSTANCE instance;
1226 }
1227 OpenedDLL;
1228
1229 /* A list thereof. */
1230 static OpenedDLL* opened_dlls = NULL;
1231 #endif
1232
1233 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1234
1235 static const char *
1236 internal_dlopen(const char *dll_name)
1237 {
1238 void *hdl;
1239 const char *errmsg;
1240 char *errmsg_copy;
1241
1242 // omitted: RTLD_NOW
1243 // see http://www.haskell.org/pipermail/cvs-ghc/2007-September/038570.html
1244 IF_DEBUG(linker,
1245 debugBelch("internal_dlopen: dll_name = '%s'\n", dll_name));
1246
1247 //-------------- Begin critical section ------------------
1248 // This critical section is necessary because dlerror() is not
1249 // required to be reentrant (see POSIX -- IEEE Std 1003.1-2008)
1250 // Also, the error message returned must be copied to preserve it
1251 // (see POSIX also)
1252
1253 ACQUIRE_LOCK(&dl_mutex);
1254 hdl = dlopen(dll_name, RTLD_LAZY | RTLD_GLOBAL);
1255
1256 errmsg = NULL;
1257 if (hdl == NULL) {
1258 /* dlopen failed; return a ptr to the error msg. */
1259 errmsg = dlerror();
1260 if (errmsg == NULL) errmsg = "addDLL: unknown error";
1261 errmsg_copy = stgMallocBytes(strlen(errmsg)+1, "addDLL");
1262 strcpy(errmsg_copy, errmsg);
1263 errmsg = errmsg_copy;
1264 }
1265 RELEASE_LOCK(&dl_mutex);
1266 //--------------- End critical section -------------------
1267
1268 return errmsg;
1269 }
1270 # endif
1271
1272 const char *
1273 addDLL( char *dll_name )
1274 {
1275 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1276 /* ------------------- ELF DLL loader ------------------- */
1277
1278 #define NMATCH 5
1279 regmatch_t match[NMATCH];
1280 const char *errmsg;
1281 FILE* fp;
1282 size_t match_length;
1283 #define MAXLINE 1000
1284 char line[MAXLINE];
1285 int result;
1286
1287 initLinker();
1288
1289 IF_DEBUG(linker, debugBelch("addDLL: dll_name = '%s'\n", dll_name));
1290 errmsg = internal_dlopen(dll_name);
1291
1292 if (errmsg == NULL) {
1293 return NULL;
1294 }
1295
1296 // GHC Trac ticket #2615
1297 // On some systems (e.g., Gentoo Linux) dynamic files (e.g. libc.so)
1298 // contain linker scripts rather than ELF-format object code. This
1299 // code handles the situation by recognizing the real object code
1300 // file name given in the linker script.
1301 //
1302 // If an "invalid ELF header" error occurs, it is assumed that the
1303 // .so file contains a linker script instead of ELF object code.
1304 // In this case, the code looks for the GROUP ( ... ) linker
1305 // directive. If one is found, the first file name inside the
1306 // parentheses is treated as the name of a dynamic library and the
1307 // code attempts to dlopen that file. If this is also unsuccessful,
1308 // an error message is returned.
1309
1310 // see if the error message is due to an invalid ELF header
1311 IF_DEBUG(linker, debugBelch("errmsg = '%s'\n", errmsg));
1312 result = regexec(&re_invalid, errmsg, (size_t) NMATCH, match, 0);
1313 IF_DEBUG(linker, debugBelch("result = %i\n", result));
1314 if (result == 0) {
1315 // success -- try to read the named file as a linker script
1316 match_length = (size_t) stg_min((match[1].rm_eo - match[1].rm_so),
1317 MAXLINE-1);
1318 strncpy(line, (errmsg+(match[1].rm_so)),match_length);
1319 line[match_length] = '\0'; // make sure string is null-terminated
1320 IF_DEBUG(linker, debugBelch ("file name = '%s'\n", line));
1321 if ((fp = fopen(line, "r")) == NULL) {
1322 return errmsg; // return original error if open fails
1323 }
1324 // try to find a GROUP ( ... ) command
1325 while (fgets(line, MAXLINE, fp) != NULL) {
1326 IF_DEBUG(linker, debugBelch("input line = %s", line));
1327 if (regexec(&re_realso, line, (size_t) NMATCH, match, 0) == 0) {
1328 // success -- try to dlopen the first named file
1329 IF_DEBUG(linker, debugBelch("match%s\n",""));
1330 line[match[1].rm_eo] = '\0';
1331 errmsg = internal_dlopen(line+match[1].rm_so);
1332 break;
1333 }
1334 // if control reaches here, no GROUP ( ... ) directive was found
1335 // and the original error message is returned to the caller
1336 }
1337 fclose(fp);
1338 }
1339 return errmsg;
1340
1341 # elif defined(OBJFORMAT_PEi386)
1342 /* ------------------- Win32 DLL loader ------------------- */
1343
1344 char* buf;
1345 OpenedDLL* o_dll;
1346 HINSTANCE instance;
1347
1348 initLinker();
1349
1350 /* debugBelch("\naddDLL; dll_name = `%s'\n", dll_name); */
1351
1352 /* See if we've already got it, and ignore if so. */
1353 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
1354 if (0 == strcmp(o_dll->name, dll_name))
1355 return NULL;
1356 }
1357
1358 /* The file name has no suffix (yet) so that we can try
1359 both foo.dll and foo.drv
1360
1361 The documentation for LoadLibrary says:
1362 If no file name extension is specified in the lpFileName
1363 parameter, the default library extension .dll is
1364 appended. However, the file name string can include a trailing
1365 point character (.) to indicate that the module name has no
1366 extension. */
1367
1368 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
1369 sprintf(buf, "%s.DLL", dll_name);
1370 instance = LoadLibrary(buf);
1371 if (instance == NULL) {
1372 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1373 // KAA: allow loading of drivers (like winspool.drv)
1374 sprintf(buf, "%s.DRV", dll_name);
1375 instance = LoadLibrary(buf);
1376 if (instance == NULL) {
1377 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1378 // #1883: allow loading of unix-style libfoo.dll DLLs
1379 sprintf(buf, "lib%s.DLL", dll_name);
1380 instance = LoadLibrary(buf);
1381 if (instance == NULL) {
1382 goto error;
1383 }
1384 }
1385 }
1386 stgFree(buf);
1387
1388 /* Add this DLL to the list of DLLs in which to search for symbols. */
1389 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
1390 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
1391 strcpy(o_dll->name, dll_name);
1392 o_dll->instance = instance;
1393 o_dll->next = opened_dlls;
1394 opened_dlls = o_dll;
1395
1396 return NULL;
1397
1398 error:
1399 stgFree(buf);
1400 sysErrorBelch(dll_name);
1401
1402 /* LoadLibrary failed; return a ptr to the error msg. */
1403 return "addDLL: could not load DLL";
1404
1405 # else
1406 barf("addDLL: not implemented on this platform");
1407 # endif
1408 }
1409
1410 /* -----------------------------------------------------------------------------
1411 * insert a stable symbol in the hash table
1412 */
1413
1414 void
1415 insertStableSymbol(char* obj_name, char* key, StgPtr p)
1416 {
1417 ghciInsertStrHashTable(obj_name, stablehash, key, getStablePtr(p));
1418 }
1419
1420
1421 /* -----------------------------------------------------------------------------
1422 * insert a symbol in the hash table
1423 */
1424 void
1425 insertSymbol(char* obj_name, char* key, void* data)
1426 {
1427 ghciInsertStrHashTable(obj_name, symhash, key, data);
1428 }
1429
1430 /* -----------------------------------------------------------------------------
1431 * lookup a symbol in the hash table
1432 */
1433 void *
1434 lookupSymbol( char *lbl )
1435 {
1436 void *val;
1437 initLinker() ;
1438 ASSERT(symhash != NULL);
1439 val = lookupStrHashTable(symhash, lbl);
1440
1441 if (val == NULL) {
1442 # if defined(OBJFORMAT_ELF)
1443 return dlsym(dl_prog_handle, lbl);
1444 # elif defined(OBJFORMAT_MACHO)
1445 # if HAVE_DLFCN_H
1446 /* On OS X 10.3 and later, we use dlsym instead of the old legacy
1447 interface.
1448
1449 HACK: On OS X, global symbols are prefixed with an underscore.
1450 However, dlsym wants us to omit the leading underscore from the
1451 symbol name. For now, we simply strip it off here (and ONLY
1452 here).
1453 */
1454 ASSERT(lbl[0] == '_');
1455 return dlsym(dl_prog_handle, lbl+1);
1456 # else
1457 if(NSIsSymbolNameDefined(lbl)) {
1458 NSSymbol symbol = NSLookupAndBindSymbol(lbl);
1459 return NSAddressOfSymbol(symbol);
1460 } else {
1461 return NULL;
1462 }
1463 # endif /* HAVE_DLFCN_H */
1464 # elif defined(OBJFORMAT_PEi386)
1465 void* sym;
1466
1467 sym = lookupSymbolInDLLs((unsigned char*)lbl);
1468 if (sym != NULL) { return sym; };
1469
1470 // Also try looking up the symbol without the @N suffix. Some
1471 // DLLs have the suffixes on their symbols, some don't.
1472 zapTrailingAtSign ( (unsigned char*)lbl );
1473 sym = lookupSymbolInDLLs((unsigned char*)lbl);
1474 if (sym != NULL) { return sym; };
1475 return NULL;
1476
1477 # else
1478 ASSERT(2+2 == 5);
1479 return NULL;
1480 # endif
1481 } else {
1482 return val;
1483 }
1484 }
1485
1486 /* -----------------------------------------------------------------------------
1487 * Debugging aid: look in GHCi's object symbol tables for symbols
1488 * within DELTA bytes of the specified address, and show their names.
1489 */
1490 #ifdef DEBUG
1491 void ghci_enquire ( char* addr );
1492
1493 void ghci_enquire ( char* addr )
1494 {
1495 int i;
1496 char* sym;
1497 char* a;
1498 const int DELTA = 64;
1499 ObjectCode* oc;
1500
1501 initLinker();
1502
1503 for (oc = objects; oc; oc = oc->next) {
1504 for (i = 0; i < oc->n_symbols; i++) {
1505 sym = oc->symbols[i];
1506 if (sym == NULL) continue;
1507 a = NULL;
1508 if (a == NULL) {
1509 a = lookupStrHashTable(symhash, sym);
1510 }
1511 if (a == NULL) {
1512 // debugBelch("ghci_enquire: can't find %s\n", sym);
1513 }
1514 else if (addr-DELTA <= a && a <= addr+DELTA) {
1515 debugBelch("%p + %3d == `%s'\n", addr, (int)(a - addr), sym);
1516 }
1517 }
1518 }
1519 }
1520 #endif
1521
1522 #ifdef USE_MMAP
1523 #define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))
1524
1525 static void *
1526 mmapForLinker (size_t bytes, nat flags, int fd)
1527 {
1528 void *map_addr = NULL;
1529 void *result;
1530 int pagesize, size;
1531 static nat fixed = 0;
1532
1533 pagesize = getpagesize();
1534 size = ROUND_UP(bytes, pagesize);
1535
1536 #if defined(x86_64_HOST_ARCH)
1537 mmap_again:
1538
1539 if (mmap_32bit_base != 0) {
1540 map_addr = mmap_32bit_base;
1541 }
1542 #endif
1543
1544 result = mmap(map_addr, size, PROT_EXEC|PROT_READ|PROT_WRITE,
1545 MAP_PRIVATE|TRY_MAP_32BIT|fixed|flags, fd, 0);
1546
1547 if (result == MAP_FAILED) {
1548 sysErrorBelch("mmap %lu bytes at %p",(lnat)size,map_addr);
1549 errorBelch("Try specifying an address with +RTS -xm<addr> -RTS");
1550 stg_exit(EXIT_FAILURE);
1551 }
1552
1553 #if defined(x86_64_HOST_ARCH)
1554 if (mmap_32bit_base != 0) {
1555 if (result == map_addr) {
1556 mmap_32bit_base = (StgWord8*)map_addr + size;
1557 } else {
1558 if ((W_)result > 0x80000000) {
1559 // oops, we were given memory over 2Gb
1560 #if defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS)
1561 // Some platforms require MAP_FIXED. This is normally
1562 // a bad idea, because MAP_FIXED will overwrite
1563 // existing mappings.
1564 munmap(result,size);
1565 fixed = MAP_FIXED;
1566 goto mmap_again;
1567 #else
1568 barf("loadObj: failed to mmap() memory below 2Gb; asked for %lu bytes at %p. Try specifying an address with +RTS -xm<addr> -RTS", size, map_addr, result);
1569 #endif
1570 } else {
1571 // hmm, we were given memory somewhere else, but it's
1572 // still under 2Gb so we can use it. Next time, ask
1573 // for memory right after the place we just got some
1574 mmap_32bit_base = (StgWord8*)result + size;
1575 }
1576 }
1577 } else {
1578 if ((W_)result > 0x80000000) {
1579 // oops, we were given memory over 2Gb
1580 // ... try allocating memory somewhere else?;
1581 debugTrace(DEBUG_linker,"MAP_32BIT didn't work; gave us %lu bytes at 0x%p", bytes, result);
1582 munmap(result, size);
1583
1584 // Set a base address and try again... (guess: 1Gb)
1585 mmap_32bit_base = (void*)0x40000000;
1586 goto mmap_again;
1587 }
1588 }
1589 #endif
1590
1591 return result;
1592 }
1593 #endif // USE_MMAP
1594
1595 /* -----------------------------------------------------------------------------
1596 * Load an obj (populate the global symbol table, but don't resolve yet)
1597 *
1598 * Returns: 1 if ok, 0 on error.
1599 */
1600 HsInt
1601 loadObj( char *path )
1602 {
1603 ObjectCode* oc;
1604 struct stat st;
1605 int r;
1606 #ifdef USE_MMAP
1607 int fd;
1608 #else
1609 FILE *f;
1610 #endif
1611 initLinker();
1612
1613 /* debugBelch("loadObj %s\n", path ); */
1614
1615 /* Check that we haven't already loaded this object.
1616 Ignore requests to load multiple times */
1617 {
1618 ObjectCode *o;
1619 int is_dup = 0;
1620 for (o = objects; o; o = o->next) {
1621 if (0 == strcmp(o->fileName, path)) {
1622 is_dup = 1;
1623 break; /* don't need to search further */
1624 }
1625 }
1626 if (is_dup) {
1627 IF_DEBUG(linker, debugBelch(
1628 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1629 "same object file twice:\n"
1630 " %s\n"
1631 "GHCi will ignore this, but be warned.\n"
1632 , path));
1633 return 1; /* success */
1634 }
1635 }
1636
1637 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
1638
1639 # if defined(OBJFORMAT_ELF)
1640 oc->formatName = "ELF";
1641 # elif defined(OBJFORMAT_PEi386)
1642 oc->formatName = "PEi386";
1643 # elif defined(OBJFORMAT_MACHO)
1644 oc->formatName = "Mach-O";
1645 # else
1646 stgFree(oc);
1647 barf("loadObj: not implemented on this platform");
1648 # endif
1649
1650 r = stat(path, &st);
1651 if (r == -1) { return 0; }
1652
1653 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1654 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1655 strcpy(oc->fileName, path);
1656
1657 oc->fileSize = st.st_size;
1658 oc->symbols = NULL;
1659 oc->sections = NULL;
1660 oc->proddables = NULL;
1661
1662 /* chain it onto the list of objects */
1663 oc->next = objects;
1664 objects = oc;
1665
1666 #ifdef USE_MMAP
1667 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1668
1669 #if defined(openbsd_HOST_OS)
1670 fd = open(path, O_RDONLY, S_IRUSR);
1671 #else
1672 fd = open(path, O_RDONLY);
1673 #endif
1674 if (fd == -1)
1675 barf("loadObj: can't open `%s'", path);
1676
1677 oc->image = mmapForLinker(oc->fileSize, 0, fd);
1678
1679 close(fd);
1680
1681 #else /* !USE_MMAP */
1682 /* load the image into memory */
1683 f = fopen(path, "rb");
1684 if (!f)
1685 barf("loadObj: can't read `%s'", path);
1686
1687 # if defined(mingw32_HOST_OS)
1688 // TODO: We would like to use allocateExec here, but allocateExec
1689 // cannot currently allocate blocks large enough.
1690 oc->image = VirtualAlloc(NULL, oc->fileSize, MEM_RESERVE | MEM_COMMIT,
1691 PAGE_EXECUTE_READWRITE);
1692 # elif defined(darwin_HOST_OS)
1693 // In a Mach-O .o file, all sections can and will be misaligned
1694 // if the total size of the headers is not a multiple of the
1695 // desired alignment. This is fine for .o files that only serve
1696 // as input for the static linker, but it's not fine for us,
1697 // as SSE (used by gcc for floating point) and Altivec require
1698 // 16-byte alignment.
1699 // We calculate the correct alignment from the header before
1700 // reading the file, and then we misalign oc->image on purpose so
1701 // that the actual sections end up aligned again.
1702 oc->misalignment = machoGetMisalignment(f);
1703 oc->image = stgMallocBytes(oc->fileSize + oc->misalignment, "loadObj(image)");
1704 oc->image += oc->misalignment;
1705 # else
1706 oc->image = stgMallocBytes(oc->fileSize, "loadObj(image)");
1707 # endif
1708
1709 {
1710 int n;
1711 n = fread ( oc->image, 1, oc->fileSize, f );
1712 if (n != oc->fileSize)
1713 barf("loadObj: error whilst reading `%s'", path);
1714 }
1715 fclose(f);
1716 #endif /* USE_MMAP */
1717
1718 # if defined(OBJFORMAT_MACHO) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1719 r = ocAllocateSymbolExtras_MachO ( oc );
1720 if (!r) { return r; }
1721 # elif defined(OBJFORMAT_ELF) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1722 r = ocAllocateSymbolExtras_ELF ( oc );
1723 if (!r) { return r; }
1724 #endif
1725
1726 /* verify the in-memory image */
1727 # if defined(OBJFORMAT_ELF)
1728 r = ocVerifyImage_ELF ( oc );
1729 # elif defined(OBJFORMAT_PEi386)
1730 r = ocVerifyImage_PEi386 ( oc );
1731 # elif defined(OBJFORMAT_MACHO)
1732 r = ocVerifyImage_MachO ( oc );
1733 # else
1734 barf("loadObj: no verify method");
1735 # endif
1736 if (!r) { return r; }
1737
1738 /* build the symbol list for this image */
1739 # if defined(OBJFORMAT_ELF)
1740 r = ocGetNames_ELF ( oc );
1741 # elif defined(OBJFORMAT_PEi386)
1742 r = ocGetNames_PEi386 ( oc );
1743 # elif defined(OBJFORMAT_MACHO)
1744 r = ocGetNames_MachO ( oc );
1745 # else
1746 barf("loadObj: no getNames method");
1747 # endif
1748 if (!r) { return r; }
1749
1750 /* loaded, but not resolved yet */
1751 oc->status = OBJECT_LOADED;
1752
1753 return 1;
1754 }
1755
1756 /* -----------------------------------------------------------------------------
1757 * resolve all the currently unlinked objects in memory
1758 *
1759 * Returns: 1 if ok, 0 on error.
1760 */
1761 HsInt
1762 resolveObjs( void )
1763 {
1764 ObjectCode *oc;
1765 int r;
1766
1767 initLinker();
1768
1769 for (oc = objects; oc; oc = oc->next) {
1770 if (oc->status != OBJECT_RESOLVED) {
1771 # if defined(OBJFORMAT_ELF)
1772 r = ocResolve_ELF ( oc );
1773 # elif defined(OBJFORMAT_PEi386)
1774 r = ocResolve_PEi386 ( oc );
1775 # elif defined(OBJFORMAT_MACHO)
1776 r = ocResolve_MachO ( oc );
1777 # else
1778 barf("resolveObjs: not implemented on this platform");
1779 # endif
1780 if (!r) { return r; }
1781 oc->status = OBJECT_RESOLVED;
1782 }
1783 }
1784 return 1;
1785 }
1786
1787 /* -----------------------------------------------------------------------------
1788 * delete an object from the pool
1789 */
1790 HsInt
1791 unloadObj( char *path )
1792 {
1793 ObjectCode *oc, *prev;
1794
1795 ASSERT(symhash != NULL);
1796 ASSERT(objects != NULL);
1797
1798 initLinker();
1799
1800 prev = NULL;
1801 for (oc = objects; oc; prev = oc, oc = oc->next) {
1802 if (!strcmp(oc->fileName,path)) {
1803
1804 /* Remove all the mappings for the symbols within this
1805 * object..
1806 */
1807 {
1808 int i;
1809 for (i = 0; i < oc->n_symbols; i++) {
1810 if (oc->symbols[i] != NULL) {
1811 removeStrHashTable(symhash, oc->symbols[i], NULL);
1812 }
1813 }
1814 }
1815
1816 if (prev == NULL) {
1817 objects = oc->next;
1818 } else {
1819 prev->next = oc->next;
1820 }
1821
1822 // We're going to leave this in place, in case there are
1823 // any pointers from the heap into it:
1824 // #ifdef mingw32_HOST_OS
1825 // VirtualFree(oc->image);
1826 // #else
1827 // stgFree(oc->image);
1828 // #endif
1829 stgFree(oc->fileName);
1830 stgFree(oc->symbols);
1831 stgFree(oc->sections);
1832 stgFree(oc);
1833 return 1;
1834 }
1835 }
1836
1837 errorBelch("unloadObj: can't find `%s' to unload", path);
1838 return 0;
1839 }
1840
1841 /* -----------------------------------------------------------------------------
1842 * Sanity checking. For each ObjectCode, maintain a list of address ranges
1843 * which may be prodded during relocation, and abort if we try and write
1844 * outside any of these.
1845 */
1846 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
1847 {
1848 ProddableBlock* pb
1849 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
1850 /* debugBelch("aPB %p %p %d\n", oc, start, size); */
1851 ASSERT(size > 0);
1852 pb->start = start;
1853 pb->size = size;
1854 pb->next = oc->proddables;
1855 oc->proddables = pb;
1856 }
1857
1858 static void checkProddableBlock ( ObjectCode* oc, void* addr )
1859 {
1860 ProddableBlock* pb;
1861 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
1862 char* s = (char*)(pb->start);
1863 char* e = s + pb->size - 1;
1864 char* a = (char*)addr;
1865 /* Assumes that the biggest fixup involves a 4-byte write. This
1866 probably needs to be changed to 8 (ie, +7) on 64-bit
1867 plats. */
1868 if (a >= s && (a+3) <= e) return;
1869 }
1870 barf("checkProddableBlock: invalid fixup in runtime linker");
1871 }
1872
1873 /* -----------------------------------------------------------------------------
1874 * Section management.
1875 */
1876 static void addSection ( ObjectCode* oc, SectionKind kind,
1877 void* start, void* end )
1878 {
1879 Section* s = stgMallocBytes(sizeof(Section), "addSection");
1880 s->start = start;
1881 s->end = end;
1882 s->kind = kind;
1883 s->next = oc->sections;
1884 oc->sections = s;
1885 /*
1886 debugBelch("addSection: %p-%p (size %d), kind %d\n",
1887 start, ((char*)end)-1, end - start + 1, kind );
1888 */
1889 }
1890
1891
1892 /* --------------------------------------------------------------------------
1893 * Symbol Extras.
1894 * This is about allocating a small chunk of memory for every symbol in the
1895 * object file. We make sure that the SymboLExtras are always "in range" of
1896 * limited-range PC-relative instructions on various platforms by allocating
1897 * them right next to the object code itself.
1898 */
1899
1900 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
1901
1902 /*
1903 ocAllocateSymbolExtras
1904
1905 Allocate additional space at the end of the object file image to make room
1906 for jump islands (powerpc, x86_64) and GOT entries (x86_64).
1907
1908 PowerPC relative branch instructions have a 24 bit displacement field.
1909 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
1910 If a particular imported symbol is outside this range, we have to redirect
1911 the jump to a short piece of new code that just loads the 32bit absolute
1912 address and jumps there.
1913 On x86_64, PC-relative jumps and PC-relative accesses to the GOT are limited
1914 to 32 bits (+-2GB).
1915
1916 This function just allocates space for one SymbolExtra for every
1917 undefined symbol in the object file. The code for the jump islands is
1918 filled in by makeSymbolExtra below.
1919 */
1920
1921 static int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first )
1922 {
1923 #ifdef USE_MMAP
1924 int pagesize, n, m;
1925 #endif
1926 int aligned;
1927 #ifndef USE_MMAP
1928 int misalignment = 0;
1929 #ifdef darwin_HOST_OS
1930 misalignment = oc->misalignment;
1931 #endif
1932 #endif
1933
1934 if( count > 0 )
1935 {
1936 // round up to the nearest 4
1937 aligned = (oc->fileSize + 3) & ~3;
1938
1939 #ifdef USE_MMAP
1940 pagesize = getpagesize();
1941 n = ROUND_UP( oc->fileSize, pagesize );
1942 m = ROUND_UP( aligned + sizeof (SymbolExtra) * count, pagesize );
1943
1944 /* we try to use spare space at the end of the last page of the
1945 * image for the jump islands, but if there isn't enough space
1946 * then we have to map some (anonymously, remembering MAP_32BIT).
1947 */
1948 if( m > n ) // we need to allocate more pages
1949 {
1950 oc->symbol_extras = mmapForLinker(sizeof(SymbolExtra) * count,
1951 MAP_ANONYMOUS, -1);
1952 }
1953 else
1954 {
1955 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1956 }
1957 #else
1958 oc->image -= misalignment;
1959 oc->image = stgReallocBytes( oc->image,
1960 misalignment +
1961 aligned + sizeof (SymbolExtra) * count,
1962 "ocAllocateSymbolExtras" );
1963 oc->image += misalignment;
1964
1965 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1966 #endif /* USE_MMAP */
1967
1968 memset( oc->symbol_extras, 0, sizeof (SymbolExtra) * count );
1969 }
1970 else
1971 oc->symbol_extras = NULL;
1972
1973 oc->first_symbol_extra = first;
1974 oc->n_symbol_extras = count;
1975
1976 return 1;
1977 }
1978
1979 static SymbolExtra* makeSymbolExtra( ObjectCode* oc,
1980 unsigned long symbolNumber,
1981 unsigned long target )
1982 {
1983 SymbolExtra *extra;
1984
1985 ASSERT( symbolNumber >= oc->first_symbol_extra
1986 && symbolNumber - oc->first_symbol_extra < oc->n_symbol_extras);
1987
1988 extra = &oc->symbol_extras[symbolNumber - oc->first_symbol_extra];
1989
1990 #ifdef powerpc_HOST_ARCH
1991 // lis r12, hi16(target)
1992 extra->jumpIsland.lis_r12 = 0x3d80;
1993 extra->jumpIsland.hi_addr = target >> 16;
1994
1995 // ori r12, r12, lo16(target)
1996 extra->jumpIsland.ori_r12_r12 = 0x618c;
1997 extra->jumpIsland.lo_addr = target & 0xffff;
1998
1999 // mtctr r12
2000 extra->jumpIsland.mtctr_r12 = 0x7d8903a6;
2001
2002 // bctr
2003 extra->jumpIsland.bctr = 0x4e800420;
2004 #endif
2005 #ifdef x86_64_HOST_ARCH
2006 // jmp *-14(%rip)
2007 static uint8_t jmp[] = { 0xFF, 0x25, 0xF2, 0xFF, 0xFF, 0xFF };
2008 extra->addr = target;
2009 memcpy(extra->jumpIsland, jmp, 6);
2010 #endif
2011
2012 return extra;
2013 }
2014
2015 #endif
2016
2017 /* --------------------------------------------------------------------------
2018 * PowerPC specifics (instruction cache flushing)
2019 * ------------------------------------------------------------------------*/
2020
2021 #ifdef powerpc_HOST_ARCH
2022 /*
2023 ocFlushInstructionCache
2024
2025 Flush the data & instruction caches.
2026 Because the PPC has split data/instruction caches, we have to
2027 do that whenever we modify code at runtime.
2028 */
2029
2030 static void ocFlushInstructionCache( ObjectCode *oc )
2031 {
2032 int n = (oc->fileSize + sizeof( SymbolExtra ) * oc->n_symbol_extras + 3) / 4;
2033 unsigned long *p = (unsigned long *) oc->image;
2034
2035 while( n-- )
2036 {
2037 __asm__ volatile ( "dcbf 0,%0\n\t"
2038 "sync\n\t"
2039 "icbi 0,%0"
2040 :
2041 : "r" (p)
2042 );
2043 p++;
2044 }
2045 __asm__ volatile ( "sync\n\t"
2046 "isync"
2047 );
2048 }
2049 #endif
2050
2051 /* --------------------------------------------------------------------------
2052 * PEi386 specifics (Win32 targets)
2053 * ------------------------------------------------------------------------*/
2054
2055 /* The information for this linker comes from
2056 Microsoft Portable Executable
2057 and Common Object File Format Specification
2058 revision 5.1 January 1998
2059 which SimonM says comes from the MS Developer Network CDs.
2060
2061 It can be found there (on older CDs), but can also be found
2062 online at:
2063
2064 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
2065
2066 (this is Rev 6.0 from February 1999).
2067
2068 Things move, so if that fails, try searching for it via
2069
2070 http://www.google.com/search?q=PE+COFF+specification
2071
2072 The ultimate reference for the PE format is the Winnt.h
2073 header file that comes with the Platform SDKs; as always,
2074 implementations will drift wrt their documentation.
2075
2076 A good background article on the PE format is Matt Pietrek's
2077 March 1994 article in Microsoft System Journal (MSJ)
2078 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
2079 Win32 Portable Executable File Format." The info in there
2080 has recently been updated in a two part article in
2081 MSDN magazine, issues Feb and March 2002,
2082 "Inside Windows: An In-Depth Look into the Win32 Portable
2083 Executable File Format"
2084
2085 John Levine's book "Linkers and Loaders" contains useful
2086 info on PE too.
2087 */
2088
2089
2090 #if defined(OBJFORMAT_PEi386)
2091
2092
2093
2094 typedef unsigned char UChar;
2095 typedef unsigned short UInt16;
2096 typedef unsigned int UInt32;
2097 typedef int Int32;
2098
2099
2100 typedef
2101 struct {
2102 UInt16 Machine;
2103 UInt16 NumberOfSections;
2104 UInt32 TimeDateStamp;
2105 UInt32 PointerToSymbolTable;
2106 UInt32 NumberOfSymbols;
2107 UInt16 SizeOfOptionalHeader;
2108 UInt16 Characteristics;
2109 }
2110 COFF_header;
2111
2112 #define sizeof_COFF_header 20
2113
2114
2115 typedef
2116 struct {
2117 UChar Name[8];
2118 UInt32 VirtualSize;
2119 UInt32 VirtualAddress;
2120 UInt32 SizeOfRawData;
2121 UInt32 PointerToRawData;
2122 UInt32 PointerToRelocations;
2123 UInt32 PointerToLinenumbers;
2124 UInt16 NumberOfRelocations;
2125 UInt16 NumberOfLineNumbers;
2126 UInt32 Characteristics;
2127 }
2128 COFF_section;
2129
2130 #define sizeof_COFF_section 40
2131
2132
2133 typedef
2134 struct {
2135 UChar Name[8];
2136 UInt32 Value;
2137 UInt16 SectionNumber;
2138 UInt16 Type;
2139 UChar StorageClass;
2140 UChar NumberOfAuxSymbols;
2141 }
2142 COFF_symbol;
2143
2144 #define sizeof_COFF_symbol 18
2145
2146
2147 typedef
2148 struct {
2149 UInt32 VirtualAddress;
2150 UInt32 SymbolTableIndex;
2151 UInt16 Type;
2152 }
2153 COFF_reloc;
2154
2155 #define sizeof_COFF_reloc 10
2156
2157
2158 /* From PE spec doc, section 3.3.2 */
2159 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
2160 windows.h -- for the same purpose, but I want to know what I'm
2161 getting, here. */
2162 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
2163 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
2164 #define MYIMAGE_FILE_DLL 0x2000
2165 #define MYIMAGE_FILE_SYSTEM 0x1000
2166 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
2167 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
2168 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
2169
2170 /* From PE spec doc, section 5.4.2 and 5.4.4 */
2171 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
2172 #define MYIMAGE_SYM_CLASS_STATIC 3
2173 #define MYIMAGE_SYM_UNDEFINED 0
2174
2175 /* From PE spec doc, section 4.1 */
2176 #define MYIMAGE_SCN_CNT_CODE 0x00000020
2177 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
2178 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
2179
2180 /* From PE spec doc, section 5.2.1 */
2181 #define MYIMAGE_REL_I386_DIR32 0x0006
2182 #define MYIMAGE_REL_I386_REL32 0x0014
2183
2184
2185 /* We use myindex to calculate array addresses, rather than
2186 simply doing the normal subscript thing. That's because
2187 some of the above structs have sizes which are not
2188 a whole number of words. GCC rounds their sizes up to a
2189 whole number of words, which means that the address calcs
2190 arising from using normal C indexing or pointer arithmetic
2191 are just plain wrong. Sigh.
2192 */
2193 static UChar *
2194 myindex ( int scale, void* base, int index )
2195 {
2196 return
2197 ((UChar*)base) + scale * index;
2198 }
2199
2200
2201 static void
2202 printName ( UChar* name, UChar* strtab )
2203 {
2204 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2205 UInt32 strtab_offset = * (UInt32*)(name+4);
2206 debugBelch("%s", strtab + strtab_offset );
2207 } else {
2208 int i;
2209 for (i = 0; i < 8; i++) {
2210 if (name[i] == 0) break;
2211 debugBelch("%c", name[i] );
2212 }
2213 }
2214 }
2215
2216
2217 static void
2218 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
2219 {
2220 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2221 UInt32 strtab_offset = * (UInt32*)(name+4);
2222 strncpy ( (char*)dst, (char*)strtab+strtab_offset, dstSize );
2223 dst[dstSize-1] = 0;
2224 } else {
2225 int i = 0;
2226 while (1) {
2227 if (i >= 8) break;
2228 if (name[i] == 0) break;
2229 dst[i] = name[i];
2230 i++;
2231 }
2232 dst[i] = 0;
2233 }
2234 }
2235
2236
2237 static UChar *
2238 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
2239 {
2240 UChar* newstr;
2241 /* If the string is longer than 8 bytes, look in the
2242 string table for it -- this will be correctly zero terminated.
2243 */
2244 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2245 UInt32 strtab_offset = * (UInt32*)(name+4);
2246 return ((UChar*)strtab) + strtab_offset;
2247 }
2248 /* Otherwise, if shorter than 8 bytes, return the original,
2249 which by defn is correctly terminated.
2250 */
2251 if (name[7]==0) return name;
2252 /* The annoying case: 8 bytes. Copy into a temporary
2253 (which is never freed ...)
2254 */
2255 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
2256 ASSERT(newstr);
2257 strncpy((char*)newstr,(char*)name,8);
2258 newstr[8] = 0;
2259 return newstr;
2260 }
2261
2262
2263 /* Just compares the short names (first 8 chars) */
2264 static COFF_section *
2265 findPEi386SectionCalled ( ObjectCode* oc, UChar* name )
2266 {
2267 int i;
2268 COFF_header* hdr
2269 = (COFF_header*)(oc->image);
2270 COFF_section* sectab
2271 = (COFF_section*) (
2272 ((UChar*)(oc->image))
2273 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2274 );
2275 for (i = 0; i < hdr->NumberOfSections; i++) {
2276 UChar* n1;
2277 UChar* n2;
2278 COFF_section* section_i
2279 = (COFF_section*)
2280 myindex ( sizeof_COFF_section, sectab, i );
2281 n1 = (UChar*) &(section_i->Name);
2282 n2 = name;
2283 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
2284 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
2285 n1[6]==n2[6] && n1[7]==n2[7])
2286 return section_i;
2287 }
2288
2289 return NULL;
2290 }
2291
2292
2293 static void
2294 zapTrailingAtSign ( UChar* sym )
2295 {
2296 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
2297 int i, j;
2298 if (sym[0] == 0) return;
2299 i = 0;
2300 while (sym[i] != 0) i++;
2301 i--;
2302 j = i;
2303 while (j > 0 && my_isdigit(sym[j])) j--;
2304 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
2305 # undef my_isdigit
2306 }
2307
2308 static void *
2309 lookupSymbolInDLLs ( UChar *lbl )
2310 {
2311 OpenedDLL* o_dll;
2312 void *sym;
2313
2314 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
2315 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
2316
2317 if (lbl[0] == '_') {
2318 /* HACK: if the name has an initial underscore, try stripping
2319 it off & look that up first. I've yet to verify whether there's
2320 a Rule that governs whether an initial '_' *should always* be
2321 stripped off when mapping from import lib name to the DLL name.
2322 */
2323 sym = GetProcAddress(o_dll->instance, (char*)(lbl+1));
2324 if (sym != NULL) {
2325 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
2326 return sym;
2327 }
2328 }
2329 sym = GetProcAddress(o_dll->instance, (char*)lbl);
2330 if (sym != NULL) {
2331 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
2332 return sym;
2333 }
2334 }
2335 return NULL;
2336 }
2337
2338
2339 static int
2340 ocVerifyImage_PEi386 ( ObjectCode* oc )
2341 {
2342 int i;
2343 UInt32 j, noRelocs;
2344 COFF_header* hdr;
2345 COFF_section* sectab;
2346 COFF_symbol* symtab;
2347 UChar* strtab;
2348 /* debugBelch("\nLOADING %s\n", oc->fileName); */
2349 hdr = (COFF_header*)(oc->image);
2350 sectab = (COFF_section*) (
2351 ((UChar*)(oc->image))
2352 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2353 );
2354 symtab = (COFF_symbol*) (
2355 ((UChar*)(oc->image))
2356 + hdr->PointerToSymbolTable
2357 );
2358 strtab = ((UChar*)symtab)
2359 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2360
2361 if (hdr->Machine != 0x14c) {
2362 errorBelch("%s: Not x86 PEi386", oc->fileName);
2363 return 0;
2364 }
2365 if (hdr->SizeOfOptionalHeader != 0) {
2366 errorBelch("%s: PEi386 with nonempty optional header", oc->fileName);
2367 return 0;
2368 }
2369 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
2370 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
2371 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
2372 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
2373 errorBelch("%s: Not a PEi386 object file", oc->fileName);
2374 return 0;
2375 }
2376 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
2377 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
2378 errorBelch("%s: Invalid PEi386 word size or endiannness: %d",
2379 oc->fileName,
2380 (int)(hdr->Characteristics));
2381 return 0;
2382 }
2383 /* If the string table size is way crazy, this might indicate that
2384 there are more than 64k relocations, despite claims to the
2385 contrary. Hence this test. */
2386 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
2387 #if 0
2388 if ( (*(UInt32*)strtab) > 600000 ) {
2389 /* Note that 600k has no special significance other than being
2390 big enough to handle the almost-2MB-sized lumps that
2391 constitute HSwin32*.o. */
2392 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
2393 return 0;
2394 }
2395 #endif
2396
2397 /* No further verification after this point; only debug printing. */
2398 i = 0;
2399 IF_DEBUG(linker, i=1);
2400 if (i == 0) return 1;
2401
2402 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
2403 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
2404 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
2405
2406 debugBelch("\n" );
2407 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
2408 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
2409 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
2410 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
2411 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
2412 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
2413 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
2414
2415 /* Print the section table. */
2416 debugBelch("\n" );
2417 for (i = 0; i < hdr->NumberOfSections; i++) {
2418 COFF_reloc* reltab;
2419 COFF_section* sectab_i
2420 = (COFF_section*)
2421 myindex ( sizeof_COFF_section, sectab, i );
2422 debugBelch(
2423 "\n"
2424 "section %d\n"
2425 " name `",
2426 i
2427 );
2428 printName ( sectab_i->Name, strtab );
2429 debugBelch(
2430 "'\n"
2431 " vsize %d\n"
2432 " vaddr %d\n"
2433 " data sz %d\n"
2434 " data off %d\n"
2435 " num rel %d\n"
2436 " off rel %d\n"
2437 " ptr raw 0x%x\n",
2438 sectab_i->VirtualSize,
2439 sectab_i->VirtualAddress,
2440 sectab_i->SizeOfRawData,
2441 sectab_i->PointerToRawData,
2442 sectab_i->NumberOfRelocations,
2443 sectab_i->PointerToRelocations,
2444 sectab_i->PointerToRawData
2445 );
2446 reltab = (COFF_reloc*) (
2447 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2448 );
2449
2450 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2451 /* If the relocation field (a short) has overflowed, the
2452 * real count can be found in the first reloc entry.
2453 *
2454 * See Section 4.1 (last para) of the PE spec (rev6.0).
2455 */
2456 COFF_reloc* rel = (COFF_reloc*)
2457 myindex ( sizeof_COFF_reloc, reltab, 0 );
2458 noRelocs = rel->VirtualAddress;
2459 j = 1;
2460 } else {
2461 noRelocs = sectab_i->NumberOfRelocations;
2462 j = 0;
2463 }
2464
2465 for (; j < noRelocs; j++) {
2466 COFF_symbol* sym;
2467 COFF_reloc* rel = (COFF_reloc*)
2468 myindex ( sizeof_COFF_reloc, reltab, j );
2469 debugBelch(
2470 " type 0x%-4x vaddr 0x%-8x name `",
2471 (UInt32)rel->Type,
2472 rel->VirtualAddress );
2473 sym = (COFF_symbol*)
2474 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
2475 /* Hmm..mysterious looking offset - what's it for? SOF */
2476 printName ( sym->Name, strtab -10 );
2477 debugBelch("'\n" );
2478 }
2479
2480 debugBelch("\n" );
2481 }
2482 debugBelch("\n" );
2483 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
2484 debugBelch("---START of string table---\n");
2485 for (i = 4; i < *(Int32*)strtab; i++) {
2486 if (strtab[i] == 0)
2487 debugBelch("\n"); else
2488 debugBelch("%c", strtab[i] );
2489 }
2490 debugBelch("--- END of string table---\n");
2491
2492 debugBelch("\n" );
2493 i = 0;
2494 while (1) {
2495 COFF_symbol* symtab_i;
2496 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2497 symtab_i = (COFF_symbol*)
2498 myindex ( sizeof_COFF_symbol, symtab, i );
2499 debugBelch(
2500 "symbol %d\n"
2501 " name `",
2502 i
2503 );
2504 printName ( symtab_i->Name, strtab );
2505 debugBelch(
2506 "'\n"
2507 " value 0x%x\n"
2508 " 1+sec# %d\n"
2509 " type 0x%x\n"
2510 " sclass 0x%x\n"
2511 " nAux %d\n",
2512 symtab_i->Value,
2513 (Int32)(symtab_i->SectionNumber),
2514 (UInt32)symtab_i->Type,
2515 (UInt32)symtab_i->StorageClass,
2516 (UInt32)symtab_i->NumberOfAuxSymbols
2517 );
2518 i += symtab_i->NumberOfAuxSymbols;
2519 i++;
2520 }
2521
2522 debugBelch("\n" );
2523 return 1;
2524 }
2525
2526
2527 static int
2528 ocGetNames_PEi386 ( ObjectCode* oc )
2529 {
2530 COFF_header* hdr;
2531 COFF_section* sectab;
2532 COFF_symbol* symtab;
2533 UChar* strtab;
2534
2535 UChar* sname;
2536 void* addr;
2537 int i;
2538
2539 hdr = (COFF_header*)(oc->image);
2540 sectab = (COFF_section*) (
2541 ((UChar*)(oc->image))
2542 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2543 );
2544 symtab = (COFF_symbol*) (
2545 ((UChar*)(oc->image))
2546 + hdr->PointerToSymbolTable
2547 );
2548 strtab = ((UChar*)(oc->image))
2549 + hdr->PointerToSymbolTable
2550 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2551
2552 /* Allocate space for any (local, anonymous) .bss sections. */
2553
2554 for (i = 0; i < hdr->NumberOfSections; i++) {
2555 UInt32 bss_sz;
2556 UChar* zspace;
2557 COFF_section* sectab_i
2558 = (COFF_section*)
2559 myindex ( sizeof_COFF_section, sectab, i );
2560 if (0 != strcmp((char*)sectab_i->Name, ".bss")) continue;
2561 /* sof 10/05: the PE spec text isn't too clear regarding what
2562 * the SizeOfRawData field is supposed to hold for object
2563 * file sections containing just uninitialized data -- for executables,
2564 * it is supposed to be zero; unclear what it's supposed to be
2565 * for object files. However, VirtualSize is guaranteed to be
2566 * zero for object files, which definitely suggests that SizeOfRawData
2567 * will be non-zero (where else would the size of this .bss section be
2568 * stored?) Looking at the COFF_section info for incoming object files,
2569 * this certainly appears to be the case.
2570 *
2571 * => I suspect we've been incorrectly handling .bss sections in (relocatable)
2572 * object files up until now. This turned out to bite us with ghc-6.4.1's use
2573 * of gcc-3.4.x, which has started to emit initially-zeroed-out local 'static'
2574 * variable decls into to the .bss section. (The specific function in Q which
2575 * triggered this is libraries/base/cbits/dirUtils.c:__hscore_getFolderPath())
2576 */
2577 if (sectab_i->VirtualSize == 0 && sectab_i->SizeOfRawData == 0) continue;
2578 /* This is a non-empty .bss section. Allocate zeroed space for
2579 it, and set its PointerToRawData field such that oc->image +
2580 PointerToRawData == addr_of_zeroed_space. */
2581 bss_sz = sectab_i->VirtualSize;
2582 if ( bss_sz < sectab_i->SizeOfRawData) { bss_sz = sectab_i->SizeOfRawData; }
2583 zspace = stgCallocBytes(1, bss_sz, "ocGetNames_PEi386(anonymous bss)");
2584 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
2585 addProddableBlock(oc, zspace, bss_sz);
2586 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
2587 }
2588
2589 /* Copy section information into the ObjectCode. */
2590
2591 for (i = 0; i < hdr->NumberOfSections; i++) {
2592 UChar* start;
2593 UChar* end;
2594 UInt32 sz;
2595
2596 SectionKind kind
2597 = SECTIONKIND_OTHER;
2598 COFF_section* sectab_i
2599 = (COFF_section*)
2600 myindex ( sizeof_COFF_section, sectab, i );
2601 IF_DEBUG(linker, debugBelch("section name = %s\n", sectab_i->Name ));
2602
2603 # if 0
2604 /* I'm sure this is the Right Way to do it. However, the
2605 alternative of testing the sectab_i->Name field seems to
2606 work ok with Cygwin.
2607 */
2608 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
2609 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
2610 kind = SECTIONKIND_CODE_OR_RODATA;
2611 # endif
2612
2613 if (0==strcmp(".text",(char*)sectab_i->Name) ||
2614 0==strcmp(".rdata",(char*)sectab_i->Name)||
2615 0==strcmp(".rodata",(char*)sectab_i->Name))
2616 kind = SECTIONKIND_CODE_OR_RODATA;
2617 if (0==strcmp(".data",(char*)sectab_i->Name) ||
2618 0==strcmp(".bss",(char*)sectab_i->Name))
2619 kind = SECTIONKIND_RWDATA;
2620
2621 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
2622 sz = sectab_i->SizeOfRawData;
2623 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
2624
2625 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
2626 end = start + sz - 1;
2627
2628 if (kind == SECTIONKIND_OTHER
2629 /* Ignore sections called which contain stabs debugging
2630 information. */
2631 && 0 != strcmp(".stab", (char*)sectab_i->Name)
2632 && 0 != strcmp(".stabstr", (char*)sectab_i->Name)
2633 /* ignore constructor section for now */
2634 && 0 != strcmp(".ctors", (char*)sectab_i->Name)
2635 /* ignore section generated from .ident */
2636 && 0!= strcmp("/4", (char*)sectab_i->Name)
2637 /* ignore unknown section that appeared in gcc 3.4.5(?) */
2638 && 0!= strcmp(".reloc", (char*)sectab_i->Name)
2639 ) {
2640 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName);
2641 return 0;
2642 }
2643
2644 if (kind != SECTIONKIND_OTHER && end >= start) {
2645 addSection(oc, kind, start, end);
2646 addProddableBlock(oc, start, end - start + 1);
2647 }
2648 }
2649
2650 /* Copy exported symbols into the ObjectCode. */
2651
2652 oc->n_symbols = hdr->NumberOfSymbols;
2653 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2654 "ocGetNames_PEi386(oc->symbols)");
2655 /* Call me paranoid; I don't care. */
2656 for (i = 0; i < oc->n_symbols; i++)
2657 oc->symbols[i] = NULL;
2658
2659 i = 0;
2660 while (1) {
2661 COFF_symbol* symtab_i;
2662 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2663 symtab_i = (COFF_symbol*)
2664 myindex ( sizeof_COFF_symbol, symtab, i );
2665
2666 addr = NULL;
2667
2668 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2669 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2670 /* This symbol is global and defined, viz, exported */
2671 /* for MYIMAGE_SYMCLASS_EXTERNAL
2672 && !MYIMAGE_SYM_UNDEFINED,
2673 the address of the symbol is:
2674 address of relevant section + offset in section
2675 */
2676 COFF_section* sectabent
2677 = (COFF_section*) myindex ( sizeof_COFF_section,
2678 sectab,
2679 symtab_i->SectionNumber-1 );
2680 addr = ((UChar*)(oc->image))
2681 + (sectabent->PointerToRawData
2682 + symtab_i->Value);
2683 }
2684 else
2685 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2686 && symtab_i->Value > 0) {
2687 /* This symbol isn't in any section at all, ie, global bss.
2688 Allocate zeroed space for it. */
2689 addr = stgCallocBytes(1, symtab_i->Value,
2690 "ocGetNames_PEi386(non-anonymous bss)");
2691 addSection(oc, SECTIONKIND_RWDATA, addr,
2692 ((UChar*)addr) + symtab_i->Value - 1);
2693 addProddableBlock(oc, addr, symtab_i->Value);
2694 /* debugBelch("BSS section at 0x%x\n", addr); */
2695 }
2696
2697 if (addr != NULL ) {
2698 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2699 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2700 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2701 ASSERT(i >= 0 && i < oc->n_symbols);
2702 /* cstring_from_COFF_symbol_name always succeeds. */
2703 oc->symbols[i] = (char*)sname;
2704 ghciInsertStrHashTable(oc->fileName, symhash, (char*)sname, addr);
2705 } else {
2706 # if 0
2707 debugBelch(
2708 "IGNORING symbol %d\n"
2709 " name `",
2710 i
2711 );
2712 printName ( symtab_i->Name, strtab );
2713 debugBelch(
2714 "'\n"
2715 " value 0x%x\n"
2716 " 1+sec# %d\n"
2717 " type 0x%x\n"
2718 " sclass 0x%x\n"
2719 " nAux %d\n",
2720 symtab_i->Value,
2721 (Int32)(symtab_i->SectionNumber),
2722 (UInt32)symtab_i->Type,
2723 (UInt32)symtab_i->StorageClass,
2724 (UInt32)symtab_i->NumberOfAuxSymbols
2725 );
2726 # endif
2727 }
2728
2729 i += symtab_i->NumberOfAuxSymbols;
2730 i++;
2731 }
2732
2733 return 1;
2734 }
2735
2736
2737 static int
2738 ocResolve_PEi386 ( ObjectCode* oc )
2739 {
2740 COFF_header* hdr;
2741 COFF_section* sectab;
2742 COFF_symbol* symtab;
2743 UChar* strtab;
2744
2745 UInt32 A;
2746 UInt32 S;
2747 UInt32* pP;
2748
2749 int i;
2750 UInt32 j, noRelocs;
2751
2752 /* ToDo: should be variable-sized? But is at least safe in the
2753 sense of buffer-overrun-proof. */
2754 UChar symbol[1000];
2755 /* debugBelch("resolving for %s\n", oc->fileName); */
2756
2757 hdr = (COFF_header*)(oc->image);
2758 sectab = (COFF_section*) (
2759 ((UChar*)(oc->image))
2760 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2761 );
2762 symtab = (COFF_symbol*) (
2763 ((UChar*)(oc->image))
2764 + hdr->PointerToSymbolTable
2765 );
2766 strtab = ((UChar*)(oc->image))
2767 + hdr->PointerToSymbolTable
2768 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2769
2770 for (i = 0; i < hdr->NumberOfSections; i++) {
2771 COFF_section* sectab_i
2772 = (COFF_section*)
2773 myindex ( sizeof_COFF_section, sectab, i );
2774 COFF_reloc* reltab
2775 = (COFF_reloc*) (
2776 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2777 );
2778
2779 /* Ignore sections called which contain stabs debugging
2780 information. */
2781 if (0 == strcmp(".stab", (char*)sectab_i->Name)
2782 || 0 == strcmp(".stabstr", (char*)sectab_i->Name)
2783 || 0 == strcmp(".ctors", (char*)sectab_i->Name))
2784 continue;
2785
2786 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2787 /* If the relocation field (a short) has overflowed, the
2788 * real count can be found in the first reloc entry.
2789 *
2790 * See Section 4.1 (last para) of the PE spec (rev6.0).
2791 *
2792 * Nov2003 update: the GNU linker still doesn't correctly
2793 * handle the generation of relocatable object files with
2794 * overflown relocations. Hence the output to warn of potential
2795 * troubles.
2796 */
2797 COFF_reloc* rel = (COFF_reloc*)
2798 myindex ( sizeof_COFF_reloc, reltab, 0 );
2799 noRelocs = rel->VirtualAddress;
2800
2801 /* 10/05: we now assume (and check for) a GNU ld that is capable
2802 * of handling object files with (>2^16) of relocs.
2803 */
2804 #if 0
2805 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2806 noRelocs);
2807 #endif
2808 j = 1;
2809 } else {
2810 noRelocs = sectab_i->NumberOfRelocations;
2811 j = 0;
2812 }
2813
2814
2815 for (; j < noRelocs; j++) {
2816 COFF_symbol* sym;
2817 COFF_reloc* reltab_j
2818 = (COFF_reloc*)
2819 myindex ( sizeof_COFF_reloc, reltab, j );
2820
2821 /* the location to patch */
2822 pP = (UInt32*)(
2823 ((UChar*)(oc->image))
2824 + (sectab_i->PointerToRawData
2825 + reltab_j->VirtualAddress
2826 - sectab_i->VirtualAddress )
2827 );
2828 /* the existing contents of pP */
2829 A = *pP;
2830 /* the symbol to connect to */
2831 sym = (COFF_symbol*)
2832 myindex ( sizeof_COFF_symbol,
2833 symtab, reltab_j->SymbolTableIndex );
2834 IF_DEBUG(linker,
2835 debugBelch(
2836 "reloc sec %2d num %3d: type 0x%-4x "
2837 "vaddr 0x%-8x name `",
2838 i, j,
2839 (UInt32)reltab_j->Type,
2840 reltab_j->VirtualAddress );
2841 printName ( sym->Name, strtab );
2842 debugBelch("'\n" ));
2843
2844 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2845 COFF_section* section_sym
2846 = findPEi386SectionCalled ( oc, sym->Name );
2847 if (!section_sym) {
2848 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2849 return 0;
2850 }
2851 S = ((UInt32)(oc->image))
2852 + (section_sym->PointerToRawData
2853 + sym->Value);
2854 } else {
2855 copyName ( sym->Name, strtab, symbol, 1000-1 );
2856 S = (UInt32) lookupSymbol( (char*)symbol );
2857 if ((void*)S != NULL) goto foundit;
2858 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
2859 return 0;
2860 foundit:;
2861 }
2862 checkProddableBlock(oc, pP);
2863 switch (reltab_j->Type) {
2864 case MYIMAGE_REL_I386_DIR32:
2865 *pP = A + S;
2866 break;
2867 case MYIMAGE_REL_I386_REL32:
2868 /* Tricky. We have to insert a displacement at
2869 pP which, when added to the PC for the _next_
2870 insn, gives the address of the target (S).
2871 Problem is to know the address of the next insn
2872 when we only know pP. We assume that this
2873 literal field is always the last in the insn,
2874 so that the address of the next insn is pP+4
2875 -- hence the constant 4.
2876 Also I don't know if A should be added, but so
2877 far it has always been zero.
2878
2879 SOF 05/2005: 'A' (old contents of *pP) have been observed
2880 to contain values other than zero (the 'wx' object file
2881 that came with wxhaskell-0.9.4; dunno how it was compiled..).
2882 So, add displacement to old value instead of asserting
2883 A to be zero. Fixes wxhaskell-related crashes, and no other
2884 ill effects have been observed.
2885
2886 Update: the reason why we're seeing these more elaborate
2887 relocations is due to a switch in how the NCG compiles SRTs
2888 and offsets to them from info tables. SRTs live in .(ro)data,
2889 while info tables live in .text, causing GAS to emit REL32/DISP32
2890 relocations with non-zero values. Adding the displacement is
2891 the right thing to do.
2892 */
2893 *pP = S - ((UInt32)pP) - 4 + A;
2894 break;
2895 default:
2896 debugBelch("%s: unhandled PEi386 relocation type %d",
2897 oc->fileName, reltab_j->Type);
2898 return 0;
2899 }
2900
2901 }
2902 }
2903
2904 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2905 return 1;
2906 }
2907
2908 #endif /* defined(OBJFORMAT_PEi386) */
2909
2910
2911 /* --------------------------------------------------------------------------
2912 * ELF specifics
2913 * ------------------------------------------------------------------------*/
2914
2915 #if defined(OBJFORMAT_ELF)
2916
2917 #define FALSE 0
2918 #define TRUE 1
2919
2920 #if defined(sparc_HOST_ARCH)
2921 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2922 #elif defined(i386_HOST_ARCH)
2923 # define ELF_TARGET_386 /* Used inside <elf.h> */
2924 #elif defined(x86_64_HOST_ARCH)
2925 # define ELF_TARGET_X64_64
2926 # define ELF_64BIT
2927 #endif
2928
2929 #if !defined(openbsd_HOST_OS)
2930 # include <elf.h>
2931 #else
2932 /* openbsd elf has things in different places, with diff names */
2933 # include <elf_abi.h>
2934 # include <machine/reloc.h>
2935 # define R_386_32 RELOC_32
2936 # define R_386_PC32 RELOC_PC32
2937 #endif
2938
2939 /* If elf.h doesn't define it */
2940 # ifndef R_X86_64_PC64
2941 # define R_X86_64_PC64 24
2942 # endif
2943
2944 /*
2945 * Define a set of types which can be used for both ELF32 and ELF64
2946 */
2947
2948 #ifdef ELF_64BIT
2949 #define ELFCLASS ELFCLASS64
2950 #define Elf_Addr Elf64_Addr
2951 #define Elf_Word Elf64_Word
2952 #define Elf_Sword Elf64_Sword
2953 #define Elf_Ehdr Elf64_Ehdr
2954 #define Elf_Phdr Elf64_Phdr
2955 #define Elf_Shdr Elf64_Shdr
2956 #define Elf_Sym Elf64_Sym
2957 #define Elf_Rel Elf64_Rel
2958 #define Elf_Rela Elf64_Rela
2959 #ifndef ELF_ST_TYPE
2960 #define ELF_ST_TYPE ELF64_ST_TYPE
2961 #endif
2962 #ifndef ELF_ST_BIND
2963 #define ELF_ST_BIND ELF64_ST_BIND
2964 #endif
2965 #ifndef ELF_R_TYPE
2966 #define ELF_R_TYPE ELF64_R_TYPE
2967 #endif
2968 #ifndef ELF_R_SYM
2969 #define ELF_R_SYM ELF64_R_SYM
2970 #endif
2971 #else
2972 #define ELFCLASS ELFCLASS32
2973 #define Elf_Addr Elf32_Addr
2974 #define Elf_Word Elf32_Word
2975 #define Elf_Sword Elf32_Sword
2976 #define Elf_Ehdr Elf32_Ehdr
2977 #define Elf_Phdr Elf32_Phdr
2978 #define Elf_Shdr Elf32_Shdr
2979 #define Elf_Sym Elf32_Sym
2980 #define Elf_Rel Elf32_Rel
2981 #define Elf_Rela Elf32_Rela
2982 #ifndef ELF_ST_TYPE
2983 #define ELF_ST_TYPE ELF32_ST_TYPE
2984 #endif
2985 #ifndef ELF_ST_BIND
2986 #define ELF_ST_BIND ELF32_ST_BIND
2987 #endif
2988 #ifndef ELF_R_TYPE
2989 #define ELF_R_TYPE ELF32_R_TYPE
2990 #endif
2991 #ifndef ELF_R_SYM
2992 #define ELF_R_SYM ELF32_R_SYM
2993 #endif
2994 #endif
2995
2996
2997 /*
2998 * Functions to allocate entries in dynamic sections. Currently we simply
2999 * preallocate a large number, and we don't check if a entry for the given
3000 * target already exists (a linear search is too slow). Ideally these
3001 * entries would be associated with symbols.
3002 */
3003
3004 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
3005 #define GOT_SIZE 0x20000
3006 #define FUNCTION_TABLE_SIZE 0x10000
3007 #define PLT_SIZE 0x08000
3008
3009 #ifdef ELF_NEED_GOT
3010 static Elf_Addr got[GOT_SIZE];
3011 static unsigned int gotIndex;
3012 static Elf_Addr gp_val = (Elf_Addr)got;
3013
3014 static Elf_Addr
3015 allocateGOTEntry(Elf_Addr target)
3016 {
3017 Elf_Addr *entry;
3018
3019 if (gotIndex >= GOT_SIZE)
3020 barf("Global offset table overflow");
3021
3022 entry = &got[gotIndex++];
3023 *entry = target;
3024 return (Elf_Addr)entry;
3025 }
3026 #endif
3027
3028 #ifdef ELF_FUNCTION_DESC
3029 typedef struct {
3030 Elf_Addr ip;
3031 Elf_Addr gp;
3032 } FunctionDesc;
3033
3034 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
3035 static unsigned int functionTableIndex;
3036
3037 static Elf_Addr
3038 allocateFunctionDesc(Elf_Addr target)
3039 {
3040 FunctionDesc *entry;
3041
3042 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
3043 barf("Function table overflow");
3044
3045 entry = &functionTable[functionTableIndex++];
3046 entry->ip = target;
3047 entry->gp = (Elf_Addr)gp_val;
3048 return (Elf_Addr)entry;
3049 }
3050
3051 static Elf_Addr
3052 copyFunctionDesc(Elf_Addr target)
3053 {
3054 FunctionDesc *olddesc = (FunctionDesc *)target;
3055 FunctionDesc *newdesc;
3056
3057 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
3058 newdesc->gp = olddesc->gp;
3059 return (Elf_Addr)newdesc;
3060 }
3061 #endif
3062
3063 #ifdef ELF_NEED_PLT
3064
3065 typedef struct {
3066 unsigned char code[sizeof(plt_code)];
3067 } PLTEntry;
3068
3069 static Elf_Addr
3070 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
3071 {
3072 PLTEntry *plt = (PLTEntry *)oc->plt;
3073 PLTEntry *entry;
3074
3075 if (oc->pltIndex >= PLT_SIZE)
3076 barf("Procedure table overflow");
3077
3078 entry = &plt[oc->pltIndex++];
3079 memcpy(entry->code, plt_code, sizeof(entry->code));
3080 PLT_RELOC(entry->code, target);
3081 return (Elf_Addr)entry;
3082 }
3083
3084 static unsigned int
3085 PLTSize(void)
3086 {
3087 return (PLT_SIZE * sizeof(PLTEntry));
3088 }
3089 #endif
3090
3091
3092 /*
3093 * Generic ELF functions
3094 */
3095
3096 static char *
3097 findElfSection ( void* objImage, Elf_Word sh_type )
3098 {
3099 char* ehdrC = (char*)objImage;
3100 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3101 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
3102 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3103 char* ptr = NULL;
3104 int i;
3105
3106 for (i = 0; i < ehdr->e_shnum; i++) {
3107 if (shdr[i].sh_type == sh_type
3108 /* Ignore the section header's string table. */
3109 && i != ehdr->e_shstrndx
3110 /* Ignore string tables named .stabstr, as they contain
3111 debugging info. */
3112 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3113 ) {
3114 ptr = ehdrC + shdr[i].sh_offset;
3115 break;
3116 }
3117 }
3118 return ptr;
3119 }
3120
3121 static int
3122 ocVerifyImage_ELF ( ObjectCode* oc )
3123 {
3124 Elf_Shdr* shdr;
3125 Elf_Sym* stab;
3126 int i, j, nent, nstrtab, nsymtabs;
3127 char* sh_strtab;
3128 char* strtab;
3129
3130 char* ehdrC = (char*)(oc->image);
3131 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3132
3133 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
3134 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
3135 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
3136 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
3137 errorBelch("%s: not an ELF object", oc->fileName);
3138 return 0;
3139 }
3140
3141 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
3142 errorBelch("%s: unsupported ELF format", oc->fileName);
3143 return 0;
3144 }
3145
3146 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
3147 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
3148 } else
3149 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
3150 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
3151 } else {
3152 errorBelch("%s: unknown endiannness", oc->fileName);
3153 return 0;
3154 }
3155
3156 if (ehdr->e_type != ET_REL) {
3157 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
3158 return 0;
3159 }
3160 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
3161
3162 IF_DEBUG(linker,debugBelch( "Architecture is " ));
3163 switch (ehdr->e_machine) {
3164 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
3165 #ifdef EM_SPARC32PLUS
3166 case EM_SPARC32PLUS:
3167 #endif
3168 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
3169 #ifdef EM_IA_64
3170 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
3171 #endif
3172 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
3173 #ifdef EM_X86_64
3174 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
3175 #elif defined(EM_AMD64)
3176 case EM_AMD64: IF_DEBUG(linker,debugBelch( "amd64" )); break;
3177 #endif
3178 default: IF_DEBUG(linker,debugBelch( "unknown" ));
3179 errorBelch("%s: unknown architecture (e_machine == %d)"
3180 , oc->fileName, ehdr->e_machine);
3181 return 0;
3182 }
3183
3184 IF_DEBUG(linker,debugBelch(
3185 "\nSection header table: start %ld, n_entries %d, ent_size %d\n",
3186 (long)ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
3187
3188 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
3189
3190 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3191
3192 if (ehdr->e_shstrndx == SHN_UNDEF) {
3193 errorBelch("%s: no section header string table", oc->fileName);
3194 return 0;
3195 } else {
3196 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
3197 ehdr->e_shstrndx));
3198 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3199 }
3200
3201 for (i = 0; i < ehdr->e_shnum; i++) {
3202 IF_DEBUG(linker,debugBelch("%2d: ", i ));
3203 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
3204 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
3205 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
3206 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
3207 ehdrC + shdr[i].sh_offset,
3208 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
3209
3210 if (shdr[i].sh_type == SHT_REL) {
3211 IF_DEBUG(linker,debugBelch("Rel " ));
3212 } else if (shdr[i].sh_type == SHT_RELA) {
3213 IF_DEBUG(linker,debugBelch("RelA " ));
3214 } else {
3215 IF_DEBUG(linker,debugBelch(" "));
3216 }
3217 if (sh_strtab) {
3218 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
3219 }
3220 }
3221
3222 IF_DEBUG(linker,debugBelch( "\nString tables" ));
3223 strtab = NULL;
3224 nstrtab = 0;
3225 for (i = 0; i < ehdr->e_shnum; i++) {
3226 if (shdr[i].sh_type == SHT_STRTAB
3227 /* Ignore the section header's string table. */
3228 && i != ehdr->e_shstrndx
3229 /* Ignore string tables named .stabstr, as they contain
3230 debugging info. */
3231 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3232 ) {
3233 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
3234 strtab = ehdrC + shdr[i].sh_offset;
3235 nstrtab++;
3236 }
3237 }
3238 if (nstrtab != 1) {
3239 errorBelch("%s: no string tables, or too many", oc->fileName);
3240 return 0;
3241 }
3242
3243 nsymtabs = 0;
3244 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
3245 for (i = 0; i < ehdr->e_shnum; i++) {
3246 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3247 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
3248 nsymtabs++;
3249 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3250 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3251 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%ld rem)\n",
3252 nent,
3253 (long)shdr[i].sh_size % sizeof(Elf_Sym)
3254 ));
3255 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
3256 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
3257 return 0;
3258 }
3259 for (j = 0; j < nent; j++) {
3260 IF_DEBUG(linker,debugBelch(" %2d ", j ));
3261 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
3262 (int)stab[j].st_shndx,
3263 (int)stab[j].st_size,
3264 (char*)stab[j].st_value ));
3265
3266 IF_DEBUG(linker,debugBelch("type=" ));
3267 switch (ELF_ST_TYPE(stab[j].st_info)) {
3268 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
3269 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
3270 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
3271 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
3272 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
3273 default: IF_DEBUG(linker,debugBelch("? " )); break;
3274 }
3275 IF_DEBUG(linker,debugBelch(" " ));
3276
3277 IF_DEBUG(linker,debugBelch("bind=" ));
3278 switch (ELF_ST_BIND(stab[j].st_info)) {
3279 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
3280 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
3281 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
3282 default: IF_DEBUG(linker,debugBelch("? " )); break;
3283 }
3284 IF_DEBUG(linker,debugBelch(" " ));
3285
3286 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
3287 }
3288 }
3289
3290 if (nsymtabs == 0) {
3291 errorBelch("%s: didn't find any symbol tables", oc->fileName);
3292 return 0;
3293 }
3294
3295 return 1;
3296 }
3297
3298 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
3299 {
3300 *is_bss = FALSE;
3301
3302 if (hdr->sh_type == SHT_PROGBITS
3303 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
3304 /* .text-style section */
3305 return SECTIONKIND_CODE_OR_RODATA;
3306 }
3307
3308 if (hdr->sh_type == SHT_PROGBITS
3309 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3310 /* .data-style section */
3311 return SECTIONKIND_RWDATA;
3312 }
3313
3314 if (hdr->sh_type == SHT_PROGBITS
3315 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
3316 /* .rodata-style section */
3317 return SECTIONKIND_CODE_OR_RODATA;
3318 }
3319
3320 if (hdr->sh_type == SHT_NOBITS
3321 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3322 /* .bss-style section */
3323 *is_bss = TRUE;
3324 return SECTIONKIND_RWDATA;
3325 }
3326
3327 return SECTIONKIND_OTHER;
3328 }
3329
3330
3331 static int
3332 ocGetNames_ELF ( ObjectCode* oc )
3333 {
3334 int i, j, k, nent;
3335 Elf_Sym* stab;
3336
3337 char* ehdrC = (char*)(oc->image);
3338 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3339 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
3340 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3341
3342 ASSERT(symhash != NULL);
3343
3344 if (!strtab) {
3345 errorBelch("%s: no strtab", oc->fileName);
3346 return 0;
3347 }
3348
3349 k = 0;
3350 for (i = 0; i < ehdr->e_shnum; i++) {
3351 /* Figure out what kind of section it is. Logic derived from
3352 Figure 1.14 ("Special Sections") of the ELF document
3353 ("Portable Formats Specification, Version 1.1"). */
3354 int is_bss = FALSE;
3355 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
3356
3357 if (is_bss && shdr[i].sh_size > 0) {
3358 /* This is a non-empty .bss section. Allocate zeroed space for
3359 it, and set its .sh_offset field such that
3360 ehdrC + .sh_offset == addr_of_zeroed_space. */
3361 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
3362 "ocGetNames_ELF(BSS)");
3363 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
3364 /*
3365 debugBelch("BSS section at 0x%x, size %d\n",
3366 zspace, shdr[i].sh_size);
3367 */
3368 }
3369
3370 /* fill in the section info */
3371 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
3372 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
3373 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
3374 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
3375 }
3376
3377 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3378
3379 /* copy stuff into this module's object symbol table */
3380 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3381 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3382
3383 oc->n_symbols = nent;
3384 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3385 "ocGetNames_ELF(oc->symbols)");
3386
3387 for (j = 0; j < nent; j++) {
3388
3389 char isLocal = FALSE; /* avoids uninit-var warning */
3390 char* ad = NULL;
3391 char* nm = strtab + stab[j].st_name;
3392 int secno = stab[j].st_shndx;
3393
3394 /* Figure out if we want to add it; if so, set ad to its
3395 address. Otherwise leave ad == NULL. */
3396
3397 if (secno == SHN_COMMON) {
3398 isLocal = FALSE;
3399 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
3400 /*
3401 debugBelch("COMMON symbol, size %d name %s\n",
3402 stab[j].st_size, nm);
3403 */
3404 /* Pointless to do addProddableBlock() for this area,
3405 since the linker should never poke around in it. */
3406 }
3407 else
3408 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
3409 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
3410 )
3411 /* and not an undefined symbol */
3412 && stab[j].st_shndx != SHN_UNDEF
3413 /* and not in a "special section" */
3414 && stab[j].st_shndx < SHN_LORESERVE
3415 &&
3416 /* and it's a not a section or string table or anything silly */
3417 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
3418 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
3419 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
3420 )
3421 ) {
3422 /* Section 0 is the undefined section, hence > and not >=. */
3423 ASSERT(secno > 0 && secno < ehdr->e_shnum);
3424 /*
3425 if (shdr[secno].sh_type == SHT_NOBITS) {
3426 debugBelch(" BSS symbol, size %d off %d name %s\n",
3427 stab[j].st_size, stab[j].st_value, nm);
3428 }
3429 */
3430 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
3431 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
3432 isLocal = TRUE;
3433 } else {
3434 #ifdef ELF_FUNCTION_DESC
3435 /* dlsym() and the initialisation table both give us function
3436 * descriptors, so to be consistent we store function descriptors
3437 * in the symbol table */
3438 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
3439 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
3440 #endif
3441 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s\n",
3442 ad, oc->fileName, nm ));
3443 isLocal = FALSE;
3444 }
3445 }
3446
3447 /* And the decision is ... */
3448
3449 if (ad != NULL) {
3450 ASSERT(nm != NULL);
3451 oc->symbols[j] = nm;
3452 /* Acquire! */
3453 if (isLocal) {
3454 /* Ignore entirely. */
3455 } else {
3456 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
3457 }
3458 } else {
3459 /* Skip. */
3460 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
3461 strtab + stab[j].st_name ));
3462 /*
3463 debugBelch(
3464 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
3465 (int)ELF_ST_BIND(stab[j].st_info),
3466 (int)ELF_ST_TYPE(stab[j].st_info),
3467 (int)stab[j].st_shndx,
3468 strtab + stab[j].st_name
3469 );
3470 */
3471 oc->symbols[j] = NULL;
3472 }
3473
3474 }
3475 }
3476
3477 return 1;
3478 }
3479
3480 /* Do ELF relocations which lack an explicit addend. All x86-linux
3481 relocations appear to be of this form. */
3482 static int
3483 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
3484 Elf_Shdr* shdr, int shnum,
3485 Elf_Sym* stab, char* strtab )
3486 {
3487 int j;
3488 char *symbol;
3489 Elf_Word* targ;
3490 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
3491 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
3492 int target_shndx = shdr[shnum].sh_info;
3493 int symtab_shndx = shdr[shnum].sh_link;
3494
3495 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3496 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
3497 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3498 target_shndx, symtab_shndx ));
3499
3500 /* Skip sections that we're not interested in. */
3501 {
3502 int is_bss;
3503 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
3504 if (kind == SECTIONKIND_OTHER) {
3505 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
3506 return 1;
3507 }
3508 }
3509
3510 for (j = 0; j < nent; j++) {
3511 Elf_Addr offset = rtab[j].r_offset;
3512 Elf_Addr info = rtab[j].r_info;
3513
3514 Elf_Addr P = ((Elf_Addr)targ) + offset;
3515 Elf_Word* pP = (Elf_Word*)P;
3516 Elf_Addr A = *pP;
3517 Elf_Addr S;
3518 void* S_tmp;
3519 Elf_Addr value;
3520 StgStablePtr stablePtr;
3521 StgPtr stableVal;
3522
3523 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
3524 j, (void*)offset, (void*)info ));
3525 if (!info) {
3526 IF_DEBUG(linker,debugBelch( " ZERO" ));
3527 S = 0;
3528 } else {
3529 Elf_Sym sym = stab[ELF_R_SYM(info)];
3530 /* First see if it is a local symbol. */
3531 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3532 /* Yes, so we can get the address directly from the ELF symbol
3533 table. */
3534 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3535 S = (Elf_Addr)
3536 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3537 + stab[ELF_R_SYM(info)].st_value);
3538
3539 } else {
3540 symbol = strtab + sym.st_name;
3541 stablePtr = (StgStablePtr)lookupHashTable(stablehash, (StgWord)symbol);
3542 if (NULL == stablePtr) {
3543 /* No, so look up the name in our global table. */
3544 S_tmp = lookupSymbol( symbol );
3545 S = (Elf_Addr)S_tmp;
3546 } else {
3547 stableVal = deRefStablePtr( stablePtr );
3548 S_tmp = stableVal;
3549 S = (Elf_Addr)S_tmp;
3550 }
3551 }
3552 if (!S) {
3553 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3554 return 0;
3555 }
3556 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
3557 }
3558
3559 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
3560 (void*)P, (void*)S, (void*)A ));
3561 checkProddableBlock ( oc, pP );
3562
3563 value = S + A;
3564
3565 switch (ELF_R_TYPE(info)) {
3566 # ifdef i386_HOST_ARCH
3567 case R_386_32: *pP = value; break;
3568 case R_386_PC32: *pP = value - P; break;
3569 # endif
3570 default:
3571 errorBelch("%s: unhandled ELF relocation(Rel) type %lu\n",
3572 oc->fileName, (lnat)ELF_R_TYPE(info));
3573 return 0;
3574 }
3575
3576 }
3577 return 1;
3578 }
3579
3580 /* Do ELF relocations for which explicit addends are supplied.
3581 sparc-solaris relocations appear to be of this form. */
3582 static int
3583 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
3584 Elf_Shdr* shdr, int shnum,
3585 Elf_Sym* stab, char* strtab )
3586 {
3587 int j;
3588 char *symbol = NULL;
3589 Elf_Addr targ;
3590 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
3591 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
3592 int target_shndx = shdr[shnum].sh_info;
3593 int symtab_shndx = shdr[shnum].sh_link;
3594
3595 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3596 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
3597 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3598 target_shndx, symtab_shndx ));
3599
3600 for (j = 0; j < nent; j++) {
3601 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3602 /* This #ifdef only serves to avoid unused-var warnings. */
3603 Elf_Addr offset = rtab[j].r_offset;
3604 Elf_Addr P = targ + offset;
3605 #endif
3606 Elf_Addr info = rtab[j].r_info;
3607 Elf_Addr A = rtab[j].r_addend;
3608 Elf_Addr S;
3609 void* S_tmp;
3610 Elf_Addr value;
3611 # if defined(sparc_HOST_ARCH)
3612 Elf_Word* pP = (Elf_Word*)P;
3613 Elf_Word w1, w2;
3614 # elif defined(powerpc_HOST_ARCH)
3615 Elf_Sword delta;
3616 # endif
3617
3618 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
3619 j, (void*)offset, (void*)info,
3620 (void*)A ));
3621 if (!info) {
3622 IF_DEBUG(linker,debugBelch( " ZERO" ));
3623 S = 0;
3624 } else {
3625 Elf_Sym sym = stab[ELF_R_SYM(info)];
3626 /* First see if it is a local symbol. */
3627 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3628 /* Yes, so we can get the address directly from the ELF symbol
3629 table. */
3630 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3631 S = (Elf_Addr)
3632 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3633 + stab[ELF_R_SYM(info)].st_value);
3634 #ifdef ELF_FUNCTION_DESC
3635 /* Make a function descriptor for this function */
3636 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
3637 S = allocateFunctionDesc(S + A);
3638 A = 0;
3639 }
3640 #endif
3641 } else {
3642 /* No, so look up the name in our global table. */
3643 symbol = strtab + sym.st_name;
3644 S_tmp = lookupSymbol( symbol );
3645 S = (Elf_Addr)S_tmp;
3646
3647 #ifdef ELF_FUNCTION_DESC
3648 /* If a function, already a function descriptor - we would
3649 have to copy it to add an offset. */
3650 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3651 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3652 #endif
3653 }
3654 if (!S) {
3655 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3656 return 0;
3657 }
3658 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3659 }
3660
3661 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3662 (void*)P, (void*)S, (void*)A ));
3663 /* checkProddableBlock ( oc, (void*)P ); */
3664
3665 value = S + A;
3666
3667 switch (ELF_R_TYPE(info)) {
3668 # if defined(sparc_HOST_ARCH)
3669 case R_SPARC_WDISP30:
3670 w1 = *pP & 0xC0000000;
3671 w2 = (Elf_Word)((value - P) >> 2);
3672 ASSERT((w2 & 0xC0000000) == 0);
3673 w1 |= w2;
3674 *pP = w1;
3675 break;
3676 case R_SPARC_HI22:
3677 w1 = *pP & 0xFFC00000;
3678 w2 = (Elf_Word)(value >> 10);
3679 ASSERT((w2 & 0xFFC00000) == 0);
3680 w1 |= w2;
3681 *pP = w1;
3682 break;
3683 case R_SPARC_LO10:
3684 w1 = *pP & ~0x3FF;
3685 w2 = (Elf_Word)(value & 0x3FF);
3686 ASSERT((w2 & ~0x3FF) == 0);
3687 w1 |= w2;
3688 *pP = w1;
3689 break;
3690
3691 /* According to the Sun documentation:
3692 R_SPARC_UA32
3693 This relocation type resembles R_SPARC_32, except it refers to an
3694 unaligned word. That is, the word to be relocated must be treated
3695 as four separate bytes with arbitrary alignment, not as a word
3696 aligned according to the architecture requirements.
3697 */
3698 case R_SPARC_UA32:
3699 w2 = (Elf_Word)value;
3700
3701 // SPARC doesn't do misaligned writes of 32 bit words,
3702 // so we have to do this one byte-at-a-time.
3703 char *pPc = (char*)pP;
3704 pPc[0] = (char) ((Elf_Word)(w2 & 0xff000000) >> 24);
3705 pPc[1] = (char) ((Elf_Word)(w2 & 0x00ff0000) >> 16);
3706 pPc[2] = (char) ((Elf_Word)(w2 & 0x0000ff00) >> 8);
3707 pPc[3] = (char) ((Elf_Word)(w2 & 0x000000ff));
3708 break;
3709
3710 case R_SPARC_32:
3711 w2 = (Elf_Word)value;
3712 *pP = w2;
3713 break;
3714 # elif defined(powerpc_HOST_ARCH)
3715 case R_PPC_ADDR16_LO:
3716 *(Elf32_Half*) P = value;
3717 break;
3718
3719 case R_PPC_ADDR16_HI:
3720 *(Elf32_Half*) P = value >> 16;
3721 break;
3722
3723 case R_PPC_ADDR16_HA:
3724 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3725 break;
3726
3727 case R_PPC_ADDR32:
3728 *(Elf32_Word *) P = value;
3729 break;
3730
3731 case R_PPC_REL32:
3732 *(Elf32_Word *) P = value - P;
3733 break;
3734
3735 case R_PPC_REL24:
3736 delta = value - P;
3737
3738 if( delta << 6 >> 6 != delta )
3739 {
3740 value = (Elf_Addr) (&makeSymbolExtra( oc, ELF_R_SYM(info), value )
3741 ->jumpIsland);
3742 delta = value - P;
3743
3744 if( value == 0 || delta << 6 >> 6 != delta )
3745 {
3746 barf( "Unable to make SymbolExtra for #%d",
3747 ELF_R_SYM(info) );
3748 return 0;
3749 }
3750 }
3751
3752 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3753 | (delta & 0x3fffffc);
3754 break;
3755 # endif
3756
3757 #if x86_64_HOST_ARCH
3758 case R_X86_64_64:
3759 *(Elf64_Xword *)P = value;
3760 break;
3761
3762 case R_X86_64_PC32:
3763 {
3764 StgInt64 off = value - P;
3765 if (off >= 0x7fffffffL || off < -0x80000000L) {
3766 #if X86_64_ELF_NONPIC_HACK
3767 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3768 -> jumpIsland;
3769 off = pltAddress + A - P;
3770 #else
3771 barf("R_X86_64_PC32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3772 symbol, off, oc->fileName );
3773 #endif
3774 }
3775 *(Elf64_Word *)P = (Elf64_Word)off;
3776 break;
3777 }
3778
3779 case R_X86_64_PC64:
3780 {
3781 StgInt64 off = value - P;
3782 *(Elf64_Word *)P = (Elf64_Word)off;
3783 break;
3784 }
3785
3786 case R_X86_64_32:
3787 if (value >= 0x7fffffffL) {
3788 #if X86_64_ELF_NONPIC_HACK
3789 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3790 -> jumpIsland;
3791 value = pltAddress + A;
3792 #else
3793 barf("R_X86_64_32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3794 symbol, value, oc->fileName );
3795 #endif
3796 }
3797 *(Elf64_Word *)P = (Elf64_Word)value;
3798 break;
3799
3800 case R_X86_64_32S:
3801 if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) {
3802 #if X86_64_ELF_NONPIC_HACK
3803 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3804 -> jumpIsland;
3805 value = pltAddress + A;
3806 #else
3807 barf("R_X86_64_32S relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3808 symbol, value, oc->fileName );
3809 #endif
3810 }
3811 *(Elf64_Sword *)P = (Elf64_Sword)value;
3812 break;
3813
3814 case R_X86_64_GOTPCREL:
3815 {
3816 StgInt64 gotAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)->addr;
3817 StgInt64 off = gotAddress + A - P;
3818 *(Elf64_Word *)P = (Elf64_Word)off;
3819 break;
3820 }
3821
3822 case R_X86_64_PLT32:
3823 {
3824 StgInt64 off = value - P;
3825 if (off >= 0x7fffffffL || off < -0x80000000L) {
3826 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3827 -> jumpIsland;
3828 off = pltAddress + A - P;
3829 }
3830 *(Elf64_Word *)P = (Elf64_Word)off;
3831 break;
3832 }
3833 #endif
3834
3835 default:
3836 errorBelch("%s: unhandled ELF relocation(RelA) type %lu\n",
3837 oc->fileName, (lnat)ELF_R_TYPE(info));
3838 return 0;
3839 }
3840
3841 }
3842 return 1;
3843 }
3844
3845 static int
3846 ocResolve_ELF ( ObjectCode* oc )
3847 {
3848 char *strtab;
3849 int shnum, ok;
3850 Elf_Sym* stab = NULL;
3851 char* ehdrC = (char*)(oc->image);
3852 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3853 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3854
3855 /* first find "the" symbol table */
3856 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3857
3858 /* also go find the string table */
3859 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3860
3861 if (stab == NULL || strtab == NULL) {
3862 errorBelch("%s: can't find string or symbol table", oc->fileName);
3863 return 0;
3864 }
3865
3866 /* Process the relocation sections. */
3867 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3868 if (shdr[shnum].sh_type == SHT_REL) {
3869 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3870 shnum, stab, strtab );
3871 if (!ok) return ok;
3872 }
3873 else
3874 if (shdr[shnum].sh_type == SHT_RELA) {
3875 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3876 shnum, stab, strtab );
3877 if (!ok) return ok;
3878 }
3879 }
3880
3881 #if defined(powerpc_HOST_ARCH)
3882 ocFlushInstructionCache( oc );
3883 #endif
3884
3885 return 1;
3886 }
3887
3888 /*
3889 * PowerPC & X86_64 ELF specifics
3890 */
3891
3892 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3893
3894 static int ocAllocateSymbolExtras_ELF( ObjectCode *oc )
3895 {
3896 Elf_Ehdr *ehdr;
3897 Elf_Shdr* shdr;
3898 int i;
3899
3900 ehdr = (Elf_Ehdr *) oc->image;
3901 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3902
3903 for( i = 0; i < ehdr->e_shnum; i++ )
3904 if( shdr[i].sh_type == SHT_SYMTAB )
3905 break;
3906
3907 if( i == ehdr->e_shnum )
3908 {
3909 errorBelch( "This ELF file contains no symtab" );
3910 return 0;
3911 }
3912
3913 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3914 {
3915 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3916 (int) shdr[i].sh_entsize, (int) sizeof( Elf_Sym ) );
3917
3918 return 0;
3919 }
3920
3921 return ocAllocateSymbolExtras( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3922 }
3923
3924 #endif /* powerpc */
3925
3926 #endif /* ELF */
3927
3928 /* --------------------------------------------------------------------------
3929 * Mach-O specifics
3930 * ------------------------------------------------------------------------*/
3931
3932 #if defined(OBJFORMAT_MACHO)
3933
3934 /*
3935 Support for MachO linking on Darwin/MacOS X
3936 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3937
3938 I hereby formally apologize for the hackish nature of this code.
3939 Things that need to be done:
3940 *) implement ocVerifyImage_MachO
3941 *) add still more sanity checks.
3942 */
3943
3944 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
3945 #define mach_header mach_header_64
3946 #define segment_command segment_command_64
3947 #define section section_64
3948 #define nlist nlist_64
3949 #endif
3950
3951 #ifdef powerpc_HOST_ARCH
3952 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3953 {
3954 struct mach_header *header = (struct mach_header *) oc->image;
3955 struct load_command *lc = (struct load_command *) (header + 1);
3956 unsigned i;
3957
3958 for( i = 0; i < header->ncmds; i++ )
3959 {
3960 if( lc->cmd == LC_SYMTAB )
3961 {
3962 // Find out the first and last undefined external
3963 // symbol, so we don't have to allocate too many
3964 // jump islands.
3965 struct symtab_command *symLC = (struct symtab_command *) lc;
3966 unsigned min = symLC->nsyms, max = 0;
3967 struct nlist *nlist =
3968 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
3969 : NULL;
3970 for(i=0;i<symLC->nsyms;i++)
3971 {
3972 if(nlist[i].n_type & N_STAB)
3973 ;
3974 else if(nlist[i].n_type & N_EXT)
3975 {
3976 if((nlist[i].n_type & N_TYPE) == N_UNDF
3977 && (nlist[i].n_value == 0))
3978 {
3979 if(i < min)
3980 min = i;
3981 if(i > max)
3982 max = i;
3983 }
3984 }
3985 }
3986 if(max >= min)
3987 return ocAllocateSymbolExtras(oc, max - min + 1, min);
3988
3989 break;
3990 }
3991
3992 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
3993 }
3994 return ocAllocateSymbolExtras(oc,0,0);
3995 }
3996 #endif
3997 #ifdef x86_64_HOST_ARCH
3998 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3999 {
4000 struct mach_header *header = (struct mach_header *) oc->image;
4001 struct load_command *lc = (struct load_command *) (header + 1);
4002 unsigned i;
4003
4004 for( i = 0; i < header->ncmds; i++ )
4005 {
4006 if( lc->cmd == LC_SYMTAB )
4007 {
4008 // Just allocate one entry for every symbol
4009 struct symtab_command *symLC = (struct symtab_command *) lc;
4010
4011 return ocAllocateSymbolExtras(oc, symLC->nsyms, 0);
4012 }
4013
4014 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
4015 }