Enable -prof -threaded (#886)
[ghc.git] / rts / RtsStartup.c
1 /* -----------------------------------------------------------------------------
2 *
3 * (c) The GHC Team, 1998-2002
4 *
5 * Main function for a standalone Haskell program.
6 *
7 * ---------------------------------------------------------------------------*/
8
9 // PAPI uses caddr_t, which is not POSIX
10 // #include "PosixSource.h"
11
12 #include "Rts.h"
13 #include "RtsAPI.h"
14 #include "RtsUtils.h"
15 #include "RtsFlags.h"
16 #include "OSThreads.h"
17 #include "Schedule.h" /* initScheduler */
18 #include "Stats.h" /* initStats */
19 #include "STM.h" /* initSTM */
20 #include "Signals.h"
21 #include "RtsSignals.h"
22 #include "ThrIOManager.h"
23 #include "Timer.h" /* startTimer, stopTimer */
24 #include "Weak.h"
25 #include "Ticky.h"
26 #include "StgRun.h"
27 #include "Prelude.h" /* fixupRTStoPreludeRefs */
28 #include "HsFFI.h"
29 #include "Linker.h"
30 #include "ThreadLabels.h"
31 #include "BlockAlloc.h"
32 #include "Trace.h"
33 #include "RtsTypeable.h"
34 #include "Stable.h"
35 #include "Hpc.h"
36 #include "FileLock.h"
37
38 #if defined(RTS_GTK_FRONTPANEL)
39 #include "FrontPanel.h"
40 #endif
41
42 # include "Profiling.h"
43
44 #if defined(PROFILING)
45 # include "ProfHeap.h"
46 # include "RetainerProfile.h"
47 #endif
48
49 #if defined(GRAN)
50 # include "GranSimRts.h"
51 #endif
52
53 #if defined(GRAN) || defined(PAR)
54 # include "ParallelRts.h"
55 #endif
56
57 #if defined(PAR)
58 # include "Parallel.h"
59 # include "LLC.h"
60 #endif
61
62 #if defined(mingw32_HOST_OS) && !defined(THREADED_RTS)
63 #include "win32/AsyncIO.h"
64 #endif
65
66 #include <stdlib.h>
67
68 #ifdef HAVE_TERMIOS_H
69 #include <termios.h>
70 #endif
71 #ifdef HAVE_SIGNAL_H
72 #include <signal.h>
73 #endif
74
75 #if USE_PAPI
76 #include "Papi.h"
77 #endif
78
79 // Count of how many outstanding hs_init()s there have been.
80 static int hs_init_count = 0;
81
82 // Here we save the terminal settings on the standard file
83 // descriptors, if we need to change them (eg. to support NoBuffering
84 // input).
85 static void *saved_termios[3] = {NULL,NULL,NULL};
86
87 void*
88 __hscore_get_saved_termios(int fd)
89 {
90 return (0 <= fd && fd < (int)(sizeof(saved_termios) / sizeof(*saved_termios))) ?
91 saved_termios[fd] : NULL;
92 }
93
94 void
95 __hscore_set_saved_termios(int fd, void* ts)
96 {
97 if (0 <= fd && fd < (int)(sizeof(saved_termios) / sizeof(*saved_termios))) {
98 saved_termios[fd] = ts;
99 }
100 }
101
102 /* -----------------------------------------------------------------------------
103 Initialise floating point unit on x86 (currently disabled. why?)
104 (see comment in ghc/compiler/nativeGen/MachInstrs.lhs).
105 -------------------------------------------------------------------------- */
106
107 #define X86_INIT_FPU 0
108
109 #if X86_INIT_FPU
110 static void
111 x86_init_fpu ( void )
112 {
113 __volatile unsigned short int fpu_cw;
114
115 // Grab the control word
116 __asm __volatile ("fnstcw %0" : "=m" (fpu_cw));
117
118 #if 0
119 printf("fpu_cw: %x\n", fpu_cw);
120 #endif
121
122 // Set bits 8-9 to 10 (64-bit precision).
123 fpu_cw = (fpu_cw & 0xfcff) | 0x0200;
124
125 // Store the new control word back
126 __asm __volatile ("fldcw %0" : : "m" (fpu_cw));
127 }
128 #endif
129
130 /* -----------------------------------------------------------------------------
131 Starting up the RTS
132 -------------------------------------------------------------------------- */
133
134 void
135 hs_init(int *argc, char **argv[])
136 {
137 hs_init_count++;
138 if (hs_init_count > 1) {
139 // second and subsequent inits are ignored
140 return;
141 }
142
143 #if defined(DEBUG)
144 /* Start off by initialising the allocator debugging so we can
145 * use it anywhere */
146 initAllocator();
147 #endif
148
149 /* Next we do is grab the start time...just in case we're
150 * collecting timing statistics.
151 */
152 stat_startInit();
153
154 #ifdef PAR
155 /*
156 * The parallel system needs to be initialised and synchronised before
157 * the program is run.
158 */
159 startupParallelSystem(argv);
160
161 if (*argv[0] == '-') { /* Strip off mainPE flag argument */
162 argv++;
163 argc--;
164 }
165
166 argv[1] = argv[0]; /* ignore the nPEs argument */
167 argv++; argc--;
168 #endif
169
170 /* Initialise the performance tracking library */
171 #ifdef USE_PAPI
172 {
173 int ver;
174 if ((ver = PAPI_library_init(PAPI_VER_CURRENT)) != PAPI_VER_CURRENT) {
175 if (ver > 0) {
176 errorBelch("PAPI_library_init: wrong version: %x", ver);
177 stg_exit(EXIT_FAILURE);
178 } else {
179 sysErrorBelch("PAPI_library_init");
180 stg_exit(EXIT_FAILURE);
181 }
182 }
183 }
184 #ifdef THREADED_RTS
185 {
186 int err;
187 if ((err = PAPI_thread_init(osThreadId)) < 0) {
188 barf("PAPI_thread_init: %d",err);
189 }
190 }
191 #endif
192 #endif
193
194 /* Set the RTS flags to default values. */
195
196 initRtsFlagsDefaults();
197
198 /* Call the user hook to reset defaults, if present */
199 defaultsHook();
200
201 /* Parse the flags, separating the RTS flags from the programs args */
202 if (argc != NULL && argv != NULL) {
203 setFullProgArgv(*argc,*argv);
204 setupRtsFlags(argc, *argv, &rts_argc, rts_argv);
205 setProgArgv(*argc,*argv);
206 }
207
208 /* initTracing must be after setupRtsFlags() */
209 initTracing();
210
211 #if defined(PAR)
212 /* NB: this really must be done after processing the RTS flags */
213 IF_PAR_DEBUG(verbose,
214 debugBelch("==== Synchronising system (%d PEs)\n", nPEs));
215 synchroniseSystem(); // calls initParallelSystem etc
216 #endif /* PAR */
217
218 /* initialise scheduler data structures (needs to be done before
219 * initStorage()).
220 */
221 initScheduler();
222
223 #if defined(GRAN)
224 /* And start GranSim profiling if required: */
225 if (RtsFlags.GranFlags.GranSimStats.Full)
226 init_gr_simulation(rts_argc, rts_argv, prog_argc, prog_argv);
227 #elif defined(PAR)
228 /* And start GUM profiling if required: */
229 if (RtsFlags.ParFlags.ParStats.Full)
230 init_gr_simulation(rts_argc, rts_argv, prog_argc, prog_argv);
231 #endif /* PAR || GRAN */
232
233 /* initialize the storage manager */
234 initStorage();
235
236 /* initialise the stable pointer table */
237 initStablePtrTable();
238
239 /* initialise the shared Typeable store */
240 initTypeableStore();
241
242 /* initialise file locking, if necessary */
243 #if !defined(mingw32_HOST_OS)
244 initFileLocking();
245 #endif
246
247 #if defined(DEBUG)
248 /* initialise thread label table (tso->char*) */
249 initThreadLabelTable();
250 #endif
251
252 initProfiling1();
253
254 /* start the virtual timer 'subsystem'. */
255 initTimer();
256 startTimer();
257
258 /* Initialise the stats department */
259 initStats();
260
261 #if defined(RTS_USER_SIGNALS)
262 if (RtsFlags.MiscFlags.install_signal_handlers) {
263 /* Initialise the user signal handler set */
264 initUserSignals();
265 /* Set up handler to run on SIGINT, etc. */
266 initDefaultHandlers();
267 }
268 #endif
269
270 #if defined(mingw32_HOST_OS) && !defined(THREADED_RTS)
271 startupAsyncIO();
272 #endif
273
274 #ifdef RTS_GTK_FRONTPANEL
275 if (RtsFlags.GcFlags.frontpanel) {
276 initFrontPanel();
277 }
278 #endif
279
280 #if X86_INIT_FPU
281 x86_init_fpu();
282 #endif
283
284 /* Record initialization times */
285 stat_endInit();
286 }
287
288 // Compatibility interface
289 void
290 startupHaskell(int argc, char *argv[], void (*init_root)(void))
291 {
292 hs_init(&argc, &argv);
293 if(init_root)
294 hs_add_root(init_root);
295 }
296
297
298 /* -----------------------------------------------------------------------------
299 Per-module initialisation
300
301 This process traverses all the compiled modules in the program
302 starting with "Main", and performing per-module initialisation for
303 each one.
304
305 So far, two things happen at initialisation time:
306
307 - we register stable names for each foreign-exported function
308 in that module. This prevents foreign-exported entities, and
309 things they depend on, from being garbage collected.
310
311 - we supply a unique integer to each statically declared cost
312 centre and cost centre stack in the program.
313
314 The code generator inserts a small function "__stginit_<module>" in each
315 module and calls the registration functions in each of the modules it
316 imports.
317
318 The init* functions are compiled in the same way as STG code,
319 i.e. without normal C call/return conventions. Hence we must use
320 StgRun to call this stuff.
321 -------------------------------------------------------------------------- */
322
323 /* The init functions use an explicit stack...
324 */
325 #define INIT_STACK_BLOCKS 4
326 static F_ *init_stack = NULL;
327
328 void
329 hs_add_root(void (*init_root)(void))
330 {
331 bdescr *bd;
332 nat init_sp;
333 Capability *cap = &MainCapability;
334
335 if (hs_init_count <= 0) {
336 barf("hs_add_root() must be called after hs_init()");
337 }
338
339 /* The initialisation stack grows downward, with sp pointing
340 to the last occupied word */
341 init_sp = INIT_STACK_BLOCKS*BLOCK_SIZE_W;
342 bd = allocGroup_lock(INIT_STACK_BLOCKS);
343 init_stack = (F_ *)bd->start;
344 init_stack[--init_sp] = (F_)stg_init_finish;
345 if (init_root != NULL) {
346 init_stack[--init_sp] = (F_)init_root;
347 }
348
349 cap->r.rSp = (P_)(init_stack + init_sp);
350 StgRun((StgFunPtr)stg_init, &cap->r);
351
352 freeGroup_lock(bd);
353
354 startupHpc();
355
356 // This must be done after module initialisation.
357 // ToDo: make this work in the presence of multiple hs_add_root()s.
358 initProfiling2();
359
360 // ditto.
361 #if defined(THREADED_RTS)
362 ioManagerStart();
363 #endif
364 }
365
366 /* ----------------------------------------------------------------------------
367 * Shutting down the RTS
368 *
369 * The wait_foreign parameter means:
370 * True ==> wait for any threads doing foreign calls now.
371 * False ==> threads doing foreign calls may return in the
372 * future, but will immediately block on a mutex.
373 * (capability->lock).
374 *
375 * If this RTS is a DLL that we're about to unload, then you want
376 * safe=True, otherwise the thread might return to code that has been
377 * unloaded. If this is a standalone program that is about to exit,
378 * then you can get away with safe=False, which is better because we
379 * won't hang on exit if there is a blocked foreign call outstanding.
380 *
381 ------------------------------------------------------------------------- */
382
383 static void
384 hs_exit_(rtsBool wait_foreign)
385 {
386 if (hs_init_count <= 0) {
387 errorBelch("warning: too many hs_exit()s");
388 return;
389 }
390 hs_init_count--;
391 if (hs_init_count > 0) {
392 // ignore until it's the last one
393 return;
394 }
395
396 /* start timing the shutdown */
397 stat_startExit();
398
399 #if defined(RTS_USER_SIGNALS)
400 if (RtsFlags.MiscFlags.install_signal_handlers) {
401 freeSignalHandlers();
402 }
403 #endif
404
405 #if defined(THREADED_RTS)
406 ioManagerDie();
407 #endif
408
409 /* stop all running tasks */
410 exitScheduler(wait_foreign);
411
412 #if defined(GRAN)
413 /* end_gr_simulation prints global stats if requested -- HWL */
414 if (!RtsFlags.GranFlags.GranSimStats.Suppressed)
415 end_gr_simulation();
416 #endif
417
418 /* stop the ticker */
419 stopTimer();
420 exitTimer();
421
422 /* reset the standard file descriptors to blocking mode */
423 resetNonBlockingFd(0);
424 resetNonBlockingFd(1);
425 resetNonBlockingFd(2);
426
427 #if HAVE_TERMIOS_H
428 // Reset the terminal settings on the standard file descriptors,
429 // if we changed them. See System.Posix.Internals.tcSetAttr for
430 // more details, including the reason we termporarily disable
431 // SIGTTOU here.
432 {
433 int fd;
434 sigset_t sigset, old_sigset;
435 sigemptyset(&sigset);
436 sigaddset(&sigset, SIGTTOU);
437 sigprocmask(SIG_BLOCK, &sigset, &old_sigset);
438 for (fd = 0; fd <= 2; fd++) {
439 struct termios* ts = (struct termios*)__hscore_get_saved_termios(fd);
440 if (ts != NULL) {
441 tcsetattr(fd,TCSANOW,ts);
442 }
443 }
444 sigprocmask(SIG_SETMASK, &old_sigset, NULL);
445 }
446 #endif
447
448 #if defined(PAR)
449 /* controlled exit; good thread! */
450 shutdownParallelSystem(0);
451
452 /* global statistics in parallel system */
453 PAR_TICKY_PAR_END();
454 #endif
455
456 /* stop timing the shutdown, we're about to print stats */
457 stat_endExit();
458
459 /* shutdown the hpc support (if needed) */
460 exitHpc();
461
462 // clean up things from the storage manager's point of view.
463 // also outputs the stats (+RTS -s) info.
464 exitStorage();
465
466 /* free the tasks */
467 freeScheduler();
468
469 /* free shared Typeable store */
470 exitTypeableStore();
471
472 /* free file locking tables, if necessary */
473 #if !defined(mingw32_HOST_OS)
474 freeFileLocking();
475 #endif
476
477 /* free the stable pointer table */
478 exitStablePtrTable();
479
480 #if defined(DEBUG)
481 /* free the thread label table */
482 freeThreadLabelTable();
483 #endif
484
485 #ifdef RTS_GTK_FRONTPANEL
486 if (RtsFlags.GcFlags.frontpanel) {
487 stopFrontPanel();
488 }
489 #endif
490
491 #if defined(PROFILING)
492 reportCCSProfiling();
493 #endif
494
495 endProfiling();
496 freeProfiling1();
497
498 #ifdef PROFILING
499 // Originally, this was in report_ccs_profiling(). Now, retainer
500 // profiling might tack some extra stuff on to the end of this file
501 // during endProfiling().
502 fclose(prof_file);
503 #endif
504
505 #if defined(TICKY_TICKY)
506 if (RtsFlags.TickyFlags.showTickyStats) PrintTickyInfo();
507 #endif
508
509 #if defined(mingw32_HOST_OS) && !defined(THREADED_RTS)
510 shutdownAsyncIO(wait_foreign);
511 #endif
512
513 /* free hash table storage */
514 exitHashTable();
515
516 // Finally, free all our storage
517 freeStorage();
518
519 #if defined(DEBUG)
520 /* and shut down the allocator debugging */
521 shutdownAllocator();
522 #endif
523
524 }
525
526 // The real hs_exit():
527 void
528 hs_exit(void)
529 {
530 hs_exit_(rtsTrue);
531 // be safe; this might be a DLL
532 }
533
534 // Compatibility interfaces
535 void
536 shutdownHaskell(void)
537 {
538 hs_exit();
539 }
540
541 void
542 shutdownHaskellAndExit(int n)
543 {
544 if (hs_init_count == 1) {
545 OnExitHook();
546 hs_exit_(rtsFalse);
547 // we're about to exit(), no need to wait for foreign calls to return.
548 #if defined(PAR)
549 /* really exit (stg_exit() would call shutdownParallelSystem() again) */
550 exit(n);
551 #else
552 stg_exit(n);
553 #endif
554 }
555 }
556
557 /*
558 * called from STG-land to exit the program
559 */
560
561 #ifdef PAR
562 static int exit_started=rtsFalse;
563 #endif
564
565 void (*exitFn)(int) = 0;
566
567 void
568 stg_exit(int n)
569 {
570 #ifdef PAR
571 /* HACK: avoid a loop when exiting due to a stupid error */
572 if (exit_started)
573 return;
574 exit_started=rtsTrue;
575
576 IF_PAR_DEBUG(verbose, debugBelch("==-- stg_exit %d on [%x]...", n, mytid));
577 shutdownParallelSystem(n);
578 #endif
579 if (exitFn)
580 (*exitFn)(n);
581 exit(n);
582 }