2 * interrupt-handling magic
6 * This software is part of the SBCL system. See the README file for
9 * This software is derived from the CMU CL system, which was
10 * written at Carnegie Mellon University and released into the
11 * public domain. The software is in the public domain and is
12 * provided with absolutely no warranty. See the COPYING and CREDITS
13 * files for more information.
17 /* As far as I can tell, what's going on here is:
19 * In the case of most signals, when Lisp asks us to handle the
20 * signal, the outermost handler (the one actually passed to UNIX) is
21 * either interrupt_handle_now(..) or maybe_now_maybe_later(..).
22 * In that case, the Lisp-level handler is stored in interrupt_handlers[..]
23 * and interrupt_low_level_handlers[..] is cleared.
25 * However, some signals need special handling, e.g.
27 * o the SIGSEGV (for e.g. Linux) or SIGBUS (for e.g. FreeBSD) used by the
28 * garbage collector to detect violations of write protection,
29 * because some cases of such signals (e.g. GC-related violations of
30 * write protection) are handled at C level and never passed on to
31 * Lisp. For such signals, we still store any Lisp-level handler
32 * in interrupt_handlers[..], but for the outermost handle we use
33 * the value from interrupt_low_level_handlers[..], instead of the
34 * ordinary interrupt_handle_now(..) or interrupt_handle_later(..).
36 * o the SIGTRAP (Linux/Alpha) which Lisp code uses to handle breakpoints,
37 * pseudo-atomic sections, and some classes of error (e.g. "function
38 * not defined"). This never goes anywhere near the Lisp handlers at all.
39 * See runtime/alpha-arch.c and code/signal.lisp
41 * - WHN 20000728, dan 20010128 */
49 #include <sys/types.h>
50 #ifndef LISP_FEATURE_WIN32
58 #include "interrupt.h"
66 #include "genesis/fdefn.h"
67 #include "genesis/simple-fun.h"
68 #include "genesis/cons.h"
70 static void run_deferred_handler(struct interrupt_data *data, void *v_context);
71 #ifndef LISP_FEATURE_WIN32
72 static void store_signal_data_for_later (struct interrupt_data *data,
73 void *handler, int signal,
75 os_context_t *context);
78 sigaddset_deferrable(sigset_t *s)
82 sigaddset(s, SIGTERM);
83 sigaddset(s, SIGQUIT);
84 sigaddset(s, SIGPIPE);
85 sigaddset(s, SIGALRM);
87 sigaddset(s, SIGTSTP);
88 sigaddset(s, SIGCHLD);
90 #ifndef LISP_FEATURE_HPUX
91 sigaddset(s, SIGXCPU);
92 sigaddset(s, SIGXFSZ);
94 sigaddset(s, SIGVTALRM);
95 sigaddset(s, SIGPROF);
96 sigaddset(s, SIGWINCH);
98 #if !((defined(LISP_FEATURE_DARWIN) || defined(LISP_FEATURE_FREEBSD)) && defined(LISP_FEATURE_SB_THREAD))
99 sigaddset(s, SIGUSR1);
100 sigaddset(s, SIGUSR2);
103 #ifdef LISP_FEATURE_SB_THREAD
104 sigaddset(s, SIG_INTERRUPT_THREAD);
109 sigaddset_blockable(sigset_t *s)
111 sigaddset_deferrable(s);
112 #ifdef LISP_FEATURE_SB_THREAD
113 #ifdef SIG_RESUME_FROM_GC
114 sigaddset(s, SIG_RESUME_FROM_GC);
116 sigaddset(s, SIG_STOP_FOR_GC);
120 /* initialized in interrupt_init */
121 sigset_t deferrable_sigset;
122 sigset_t blockable_sigset;
126 check_blockables_blocked_or_lose(void)
128 #if !defined(LISP_FEATURE_WIN32)
129 /* Get the current sigmask, by blocking the empty set. */
130 sigset_t empty,current;
133 thread_sigmask(SIG_BLOCK, &empty, ¤t);
134 for(i = 1; i < NSIG; i++) {
135 if (sigismember(&blockable_sigset, i) && !sigismember(¤t, i))
136 lose("blockable signal %d not blocked\n",i);
142 unblock_gc_signals(void)
144 #ifdef LISP_FEATURE_SB_THREAD
147 #if defined(SIG_RESUME_FROM_GC)
148 sigaddset(&new,SIG_RESUME_FROM_GC);
150 sigaddset(&new,SIG_STOP_FOR_GC);
151 thread_sigmask(SIG_UNBLOCK,&new,0);
156 check_interrupts_enabled_or_lose(os_context_t *context)
158 struct thread *thread=arch_os_get_current_thread();
159 if (SymbolValue(INTERRUPTS_ENABLED,thread) == NIL)
160 lose("interrupts not enabled\n");
161 if (arch_pseudo_atomic_atomic(context))
162 lose ("in pseudo atomic section\n");
165 /* When we catch an internal error, should we pass it back to Lisp to
166 * be handled in a high-level way? (Early in cold init, the answer is
167 * 'no', because Lisp is still too brain-dead to handle anything.
168 * After sufficient initialization has been completed, the answer
170 boolean internal_errors_enabled = 0;
172 #ifndef LISP_FEATURE_WIN32
173 static void (*interrupt_low_level_handlers[NSIG]) (int, siginfo_t*, void*);
175 union interrupt_handler interrupt_handlers[NSIG];
177 /* At the toplevel repl we routinely call this function. The signal
178 * mask ought to be clear anyway most of the time, but may be non-zero
179 * if we were interrupted e.g. while waiting for a queue. */
182 reset_signal_mask(void)
184 #ifndef LISP_FEATURE_WIN32
187 thread_sigmask(SIG_SETMASK,&new,0);
192 block_blockable_signals(void)
194 #ifndef LISP_FEATURE_WIN32
195 thread_sigmask(SIG_BLOCK, &blockable_sigset, 0);
200 block_deferrable_signals(void)
202 #ifndef LISP_FEATURE_WIN32
203 thread_sigmask(SIG_BLOCK, &deferrable_sigset, 0);
209 * utility routines used by various signal handlers
213 build_fake_control_stack_frames(struct thread *th,os_context_t *context)
215 #ifndef LISP_FEATURE_C_STACK_IS_CONTROL_STACK
219 /* Build a fake stack frame or frames */
221 current_control_frame_pointer =
222 (lispobj *)(unsigned long)
223 (*os_context_register_addr(context, reg_CSP));
224 if ((lispobj *)(unsigned long)
225 (*os_context_register_addr(context, reg_CFP))
226 == current_control_frame_pointer) {
227 /* There is a small window during call where the callee's
228 * frame isn't built yet. */
229 if (lowtag_of(*os_context_register_addr(context, reg_CODE))
230 == FUN_POINTER_LOWTAG) {
231 /* We have called, but not built the new frame, so
232 * build it for them. */
233 current_control_frame_pointer[0] =
234 *os_context_register_addr(context, reg_OCFP);
235 current_control_frame_pointer[1] =
236 *os_context_register_addr(context, reg_LRA);
237 current_control_frame_pointer += 8;
238 /* Build our frame on top of it. */
239 oldcont = (lispobj)(*os_context_register_addr(context, reg_CFP));
242 /* We haven't yet called, build our frame as if the
243 * partial frame wasn't there. */
244 oldcont = (lispobj)(*os_context_register_addr(context, reg_OCFP));
247 /* We can't tell whether we are still in the caller if it had to
248 * allocate a stack frame due to stack arguments. */
249 /* This observation provoked some past CMUCL maintainer to ask
250 * "Can anything strange happen during return?" */
253 oldcont = (lispobj)(*os_context_register_addr(context, reg_CFP));
256 current_control_stack_pointer = current_control_frame_pointer + 8;
258 current_control_frame_pointer[0] = oldcont;
259 current_control_frame_pointer[1] = NIL;
260 current_control_frame_pointer[2] =
261 (lispobj)(*os_context_register_addr(context, reg_CODE));
265 /* Stores the context for gc to scavange and builds fake stack
268 fake_foreign_function_call(os_context_t *context)
271 struct thread *thread=arch_os_get_current_thread();
273 /* context_index incrementing must not be interrupted */
274 check_blockables_blocked_or_lose();
276 /* Get current Lisp state from context. */
278 dynamic_space_free_pointer =
279 (lispobj *)(unsigned long)
280 (*os_context_register_addr(context, reg_ALLOC));
281 /* fprintf(stderr,"dynamic_space_free_pointer: %p\n", */
282 /* dynamic_space_free_pointer); */
283 #if defined(LISP_FEATURE_ALPHA) || defined(LISP_FEATURE_MIPS)
284 if ((long)dynamic_space_free_pointer & 1) {
285 lose("dead in fake_foreign_function_call, context = %x\n", context);
288 /* why doesnt PPC and SPARC do something like this: */
289 #if defined(LISP_FEATURE_HPPA)
290 if ((long)dynamic_space_free_pointer & 4) {
291 lose("dead in fake_foreign_function_call, context = %x, d_s_f_p = %x\n", context, dynamic_space_free_pointer);
296 current_binding_stack_pointer =
297 (lispobj *)(unsigned long)
298 (*os_context_register_addr(context, reg_BSP));
301 build_fake_control_stack_frames(thread,context);
303 /* Do dynamic binding of the active interrupt context index
304 * and save the context in the context array. */
306 fixnum_value(SymbolValue(FREE_INTERRUPT_CONTEXT_INDEX,thread));
308 if (context_index >= MAX_INTERRUPTS) {
309 lose("maximum interrupt nesting depth (%d) exceeded\n", MAX_INTERRUPTS);
312 bind_variable(FREE_INTERRUPT_CONTEXT_INDEX,
313 make_fixnum(context_index + 1),thread);
315 thread->interrupt_contexts[context_index] = context;
317 #ifdef FOREIGN_FUNCTION_CALL_FLAG
318 foreign_function_call_active = 1;
322 /* blocks all blockable signals. If you are calling from a signal handler,
323 * the usual signal mask will be restored from the context when the handler
324 * finishes. Otherwise, be careful */
326 undo_fake_foreign_function_call(os_context_t *context)
328 struct thread *thread=arch_os_get_current_thread();
329 /* Block all blockable signals. */
330 block_blockable_signals();
332 #ifdef FOREIGN_FUNCTION_CALL_FLAG
333 foreign_function_call_active = 0;
336 /* Undo dynamic binding of FREE_INTERRUPT_CONTEXT_INDEX */
340 /* Put the dynamic space free pointer back into the context. */
341 *os_context_register_addr(context, reg_ALLOC) =
342 (unsigned long) dynamic_space_free_pointer
343 | (*os_context_register_addr(context, reg_ALLOC)
346 ((unsigned long)(*os_context_register_addr(context, reg_ALLOC))
348 | ((unsigned long) dynamic_space_free_pointer & LOWTAG_MASK);
353 /* a handler for the signal caused by execution of a trap opcode
354 * signalling an internal error */
356 interrupt_internal_error(os_context_t *context, boolean continuable)
360 fake_foreign_function_call(context);
362 if (!internal_errors_enabled) {
363 describe_internal_error(context);
364 /* There's no good way to recover from an internal error
365 * before the Lisp error handling mechanism is set up. */
366 lose("internal error too early in init, can't recover\n");
369 /* Allocate the SAP object while the interrupts are still
371 context_sap = alloc_sap(context);
373 #ifndef LISP_FEATURE_WIN32
374 thread_sigmask(SIG_SETMASK, os_context_sigmask_addr(context), 0);
377 SHOW("in interrupt_internal_error");
379 /* Display some rudimentary debugging information about the
380 * error, so that even if the Lisp error handler gets badly
381 * confused, we have a chance to determine what's going on. */
382 describe_internal_error(context);
384 funcall2(StaticSymbolFunction(INTERNAL_ERROR), context_sap,
385 continuable ? T : NIL);
387 undo_fake_foreign_function_call(context); /* blocks signals again */
389 arch_skip_instruction(context);
393 interrupt_handle_pending(os_context_t *context)
395 /* There are three ways we can get here. First, if an interrupt
396 * occurs within pseudo-atomic, it will be deferred, and we'll
397 * trap to here at the end of the pseudo-atomic block. Second, if
398 * the GC (in alloc()) decides that a GC is required, it will set
399 * *GC-PENDING* and pseudo-atomic-interrupted if not *GC-INHIBIT*,
400 * and alloc() is always called from within pseudo-atomic, and
401 * thus we end up here again. Third, when calling GC-ON or at the
402 * end of a WITHOUT-GCING, MAYBE-HANDLE-PENDING-GC will trap to
403 * here if there is a pending GC. Fourth, ahem, at the end of
404 * WITHOUT-INTERRUPTS (bar complications with nesting). */
406 /* Win32 only needs to handle the GC cases (for now?) */
408 struct thread *thread;
410 /* Punt if in PA section, marking it as interrupted. This can
411 * happenat least if we pick up a GC request while in a
412 * WITHOUT-GCING with an outer PA -- it is not immediately clear
413 * to me that this should/could ever happen, but better safe then
414 * sorry. --NS 2007-05-15 */
415 if (arch_pseudo_atomic_atomic(context)) {
416 arch_set_pseudo_atomic_interrupted(context);
420 thread = arch_os_get_current_thread();
422 FSHOW_SIGNAL((stderr, "/entering interrupt_handle_pending\n"));
424 check_blockables_blocked_or_lose();
426 /* If pseudo_atomic_interrupted is set then the interrupt is going
427 * to be handled now, ergo it's safe to clear it. */
428 arch_clear_pseudo_atomic_interrupted(context);
430 if (SymbolValue(GC_INHIBIT,thread)==NIL) {
431 #ifdef LISP_FEATURE_SB_THREAD
432 if (SymbolValue(STOP_FOR_GC_PENDING,thread) != NIL) {
433 /* STOP_FOR_GC_PENDING and GC_PENDING are cleared by
434 * the signal handler if it actually stops us. */
435 sig_stop_for_gc_handler(SIG_STOP_FOR_GC,NULL,context);
438 if (SymbolValue(GC_PENDING,thread) != NIL) {
439 /* GC_PENDING is cleared in SUB-GC, or if another thread
440 * is doing a gc already we will get a SIG_STOP_FOR_GC and
441 * that will clear it. */
444 check_blockables_blocked_or_lose();
447 #ifndef LISP_FEATURE_WIN32
448 /* we may be here only to do the gc stuff, if interrupts are
449 * enabled run the pending handler */
450 if (SymbolValue(INTERRUPTS_ENABLED,thread) != NIL) {
451 struct interrupt_data *data = thread->interrupt_data;
453 /* There may be no pending handler, because it was only a gc
454 * that had to be executed or because pseudo atomic triggered
455 * twice for a single interrupt. For the interested reader,
456 * that may happen if an interrupt hits after the interrupted
457 * flag is cleared but before pseudo-atomic is set and a
458 * pseudo atomic is interrupted in that interrupt. */
459 if (data->pending_handler) {
461 /* If we're here as the result of a pseudo-atomic as opposed
462 * to WITHOUT-INTERRUPTS, then INTERRUPT_PENDING is already
463 * NIL, because maybe_defer_handler sets
464 * PSEUDO_ATOMIC_INTERRUPTED only if interrupts are enabled.*/
465 SetSymbolValue(INTERRUPT_PENDING, NIL, thread);
467 /* restore the saved signal mask from the original signal (the
468 * one that interrupted us during the critical section) into the
469 * os_context for the signal we're currently in the handler for.
470 * This should ensure that when we return from the handler the
471 * blocked signals are unblocked */
472 sigcopyset(os_context_sigmask_addr(context), &data->pending_mask);
474 sigemptyset(&data->pending_mask);
475 /* This will break on sparc linux: the deferred handler really wants
476 * to be called with a void_context */
477 run_deferred_handler(data,(void *)context);
484 * the two main signal handlers:
485 * interrupt_handle_now(..)
486 * maybe_now_maybe_later(..)
488 * to which we have added interrupt_handle_now_handler(..). Why?
489 * Well, mostly because the SPARC/Linux platform doesn't quite do
490 * signals the way we want them done. The third argument in the
491 * handler isn't filled in by the kernel properly, so we fix it up
492 * ourselves in the arch_os_get_context(..) function; however, we only
493 * want to do this when we first hit the handler, and not when
494 * interrupt_handle_now(..) is being called from some other handler
495 * (when the fixup will already have been done). -- CSR, 2002-07-23
499 interrupt_handle_now(int signal, siginfo_t *info, os_context_t *context)
501 #ifdef FOREIGN_FUNCTION_CALL_FLAG
502 boolean were_in_lisp;
504 union interrupt_handler handler;
506 check_blockables_blocked_or_lose();
508 #ifndef LISP_FEATURE_WIN32
509 if (sigismember(&deferrable_sigset,signal))
510 check_interrupts_enabled_or_lose(context);
513 #if defined(LISP_FEATURE_LINUX) || defined(RESTORE_FP_CONTROL_FROM_CONTEXT)
514 /* Under Linux on some architectures, we appear to have to restore
515 the FPU control word from the context, as after the signal is
516 delivered we appear to have a null FPU control word. */
517 os_restore_fp_control(context);
520 handler = interrupt_handlers[signal];
522 if (ARE_SAME_HANDLER(handler.c, SIG_IGN)) {
526 #ifdef FOREIGN_FUNCTION_CALL_FLAG
527 were_in_lisp = !foreign_function_call_active;
531 fake_foreign_function_call(context);
534 FSHOW_SIGNAL((stderr,
535 "/entering interrupt_handle_now(%d, info, context)\n",
538 if (ARE_SAME_HANDLER(handler.c, SIG_DFL)) {
540 /* This can happen if someone tries to ignore or default one
541 * of the signals we need for runtime support, and the runtime
542 * support decides to pass on it. */
543 lose("no handler for signal %d in interrupt_handle_now(..)\n", signal);
545 } else if (lowtag_of(handler.lisp) == FUN_POINTER_LOWTAG) {
546 /* Once we've decided what to do about contexts in a
547 * return-elsewhere world (the original context will no longer
548 * be available; should we copy it or was nobody using it anyway?)
549 * then we should convert this to return-elsewhere */
551 /* CMUCL comment said "Allocate the SAPs while the interrupts
552 * are still disabled.". I (dan, 2003.08.21) assume this is
553 * because we're not in pseudoatomic and allocation shouldn't
554 * be interrupted. In which case it's no longer an issue as
555 * all our allocation from C now goes through a PA wrapper,
556 * but still, doesn't hurt.
558 * Yeah, but non-gencgc platforms don't really wrap allocation
559 * in PA. MG - 2005-08-29 */
561 lispobj info_sap,context_sap = alloc_sap(context);
562 info_sap = alloc_sap(info);
563 /* Leave deferrable signals blocked, the handler itself will
564 * allow signals again when it sees fit. */
565 #ifdef LISP_FEATURE_SB_THREAD
568 sigemptyset(&unblock);
569 sigaddset(&unblock, SIG_STOP_FOR_GC);
570 #ifdef SIG_RESUME_FROM_GC
571 sigaddset(&unblock, SIG_RESUME_FROM_GC);
573 thread_sigmask(SIG_UNBLOCK, &unblock, 0);
577 FSHOW_SIGNAL((stderr,"/calling Lisp-level handler\n"));
579 funcall3(handler.lisp,
585 FSHOW_SIGNAL((stderr,"/calling C-level handler\n"));
587 #ifndef LISP_FEATURE_WIN32
588 /* Allow signals again. */
589 thread_sigmask(SIG_SETMASK, os_context_sigmask_addr(context), 0);
591 (*handler.c)(signal, info, context);
594 #ifdef FOREIGN_FUNCTION_CALL_FLAG
598 undo_fake_foreign_function_call(context); /* block signals again */
601 FSHOW_SIGNAL((stderr,
602 "/returning from interrupt_handle_now(%d, info, context)\n",
606 /* This is called at the end of a critical section if the indications
607 * are that some signal was deferred during the section. Note that as
608 * far as C or the kernel is concerned we dealt with the signal
609 * already; we're just doing the Lisp-level processing now that we
612 run_deferred_handler(struct interrupt_data *data, void *v_context)
614 /* The pending_handler may enable interrupts and then another
615 * interrupt may hit, overwrite interrupt_data, so reset the
616 * pending handler before calling it. Trust the handler to finish
617 * with the siginfo before enabling interrupts. */
618 void (*pending_handler) (int, siginfo_t*, void*)=data->pending_handler;
620 data->pending_handler=0;
621 (*pending_handler)(data->pending_signal,&(data->pending_info), v_context);
624 #ifndef LISP_FEATURE_WIN32
626 maybe_defer_handler(void *handler, struct interrupt_data *data,
627 int signal, siginfo_t *info, os_context_t *context)
629 struct thread *thread=arch_os_get_current_thread();
631 check_blockables_blocked_or_lose();
633 if (SymbolValue(INTERRUPT_PENDING,thread) != NIL)
634 lose("interrupt already pending\n");
635 /* If interrupts are disabled then INTERRUPT_PENDING is set and
636 * not PSEDUO_ATOMIC_INTERRUPTED. This is important for a pseudo
637 * atomic section inside a WITHOUT-INTERRUPTS.
639 if (SymbolValue(INTERRUPTS_ENABLED,thread) == NIL) {
640 store_signal_data_for_later(data,handler,signal,info,context);
641 SetSymbolValue(INTERRUPT_PENDING, T,thread);
642 FSHOW_SIGNAL((stderr,
643 "/maybe_defer_handler(%x,%d): deferred\n",
644 (unsigned int)handler,signal));
647 /* a slightly confusing test. arch_pseudo_atomic_atomic() doesn't
648 * actually use its argument for anything on x86, so this branch
649 * may succeed even when context is null (gencgc alloc()) */
650 if (arch_pseudo_atomic_atomic(context)) {
651 store_signal_data_for_later(data,handler,signal,info,context);
652 arch_set_pseudo_atomic_interrupted(context);
653 FSHOW_SIGNAL((stderr,
654 "/maybe_defer_handler(%x,%d): deferred(PA)\n",
655 (unsigned int)handler,signal));
658 FSHOW_SIGNAL((stderr,
659 "/maybe_defer_handler(%x,%d): not deferred\n",
660 (unsigned int)handler,signal));
665 store_signal_data_for_later (struct interrupt_data *data, void *handler,
667 siginfo_t *info, os_context_t *context)
669 if (data->pending_handler)
670 lose("tried to overwrite pending interrupt handler %x with %x\n",
671 data->pending_handler, handler);
673 lose("tried to defer null interrupt handler\n");
674 data->pending_handler = handler;
675 data->pending_signal = signal;
677 memcpy(&(data->pending_info), info, sizeof(siginfo_t));
679 FSHOW_SIGNAL((stderr, "/store_signal_data_for_later: signal: %d\n",
683 /* the signal mask in the context (from before we were
684 * interrupted) is copied to be restored when
685 * run_deferred_handler happens. Then the usually-blocked
686 * signals are added to the mask in the context so that we are
687 * running with blocked signals when the handler returns */
688 sigcopyset(&(data->pending_mask),os_context_sigmask_addr(context));
689 sigaddset_deferrable(os_context_sigmask_addr(context));
694 maybe_now_maybe_later(int signal, siginfo_t *info, void *void_context)
696 os_context_t *context = arch_os_get_context(&void_context);
697 struct thread *thread = arch_os_get_current_thread();
698 struct interrupt_data *data = thread->interrupt_data;
700 #if defined(LISP_FEATURE_LINUX) || defined(RESTORE_FP_CONTROL_FROM_CONTEXT)
701 os_restore_fp_control(context);
704 if(!maybe_defer_handler(interrupt_handle_now,data,signal,info,context))
705 interrupt_handle_now(signal, info, context);
709 low_level_interrupt_handle_now(int signal, siginfo_t *info,
710 os_context_t *context)
712 /* No FP control fixage needed, caller has done that. */
713 check_blockables_blocked_or_lose();
714 check_interrupts_enabled_or_lose(context);
715 (*interrupt_low_level_handlers[signal])(signal, info, context);
716 /* No Darwin context fixage needed, caller does that. */
720 low_level_maybe_now_maybe_later(int signal, siginfo_t *info, void *void_context)
722 os_context_t *context = arch_os_get_context(&void_context);
723 struct thread *thread = arch_os_get_current_thread();
724 struct interrupt_data *data = thread->interrupt_data;
726 #if defined(LISP_FEATURE_LINUX) || defined(RESTORE_FP_CONTROL_FROM_CONTEXT)
727 os_restore_fp_control(context);
730 if(!maybe_defer_handler(low_level_interrupt_handle_now,data,
731 signal,info,context))
732 low_level_interrupt_handle_now(signal, info, context);
736 #ifdef LISP_FEATURE_SB_THREAD
739 sig_stop_for_gc_handler(int signal, siginfo_t *info, void *void_context)
741 os_context_t *context = arch_os_get_context(&void_context);
743 struct thread *thread=arch_os_get_current_thread();
746 if (arch_pseudo_atomic_atomic(context)) {
747 SetSymbolValue(STOP_FOR_GC_PENDING,T,thread);
748 arch_set_pseudo_atomic_interrupted(context);
749 FSHOW_SIGNAL((stderr, "sig_stop_for_gc deferred (PA)\n"));
752 else if (SymbolValue(GC_INHIBIT,thread) != NIL) {
753 SetSymbolValue(STOP_FOR_GC_PENDING,T,thread);
754 FSHOW_SIGNAL((stderr, "sig_stop_for_gc deferred (*GC-INHIBIT*)\n"));
758 /* Not PA and GC not inhibited -- we can stop now. */
760 /* need the context stored so it can have registers scavenged */
761 fake_foreign_function_call(context);
763 /* Block everything. */
765 thread_sigmask(SIG_BLOCK,&ss,0);
767 /* Not pending anymore. */
768 SetSymbolValue(GC_PENDING,NIL,thread);
769 SetSymbolValue(STOP_FOR_GC_PENDING,NIL,thread);
771 if(thread->state!=STATE_RUNNING) {
772 lose("sig_stop_for_gc_handler: wrong thread state: %ld\n",
773 fixnum_value(thread->state));
776 thread->state=STATE_SUSPENDED;
777 FSHOW_SIGNAL((stderr,"suspended\n"));
780 #if defined(SIG_RESUME_FROM_GC)
781 sigaddset(&ss,SIG_RESUME_FROM_GC);
783 sigaddset(&ss,SIG_STOP_FOR_GC);
786 /* It is possible to get SIGCONT (and probably other non-blockable
788 #ifdef SIG_RESUME_FROM_GC
791 do { sigwait(&ss, &sigret); }
792 while (sigret != SIG_RESUME_FROM_GC);
795 while (sigwaitinfo(&ss,0) != SIG_STOP_FOR_GC);
798 FSHOW_SIGNAL((stderr,"resumed\n"));
799 if(thread->state!=STATE_RUNNING) {
800 lose("sig_stop_for_gc_handler: wrong thread state on wakeup: %ld\n",
801 fixnum_value(thread->state));
804 undo_fake_foreign_function_call(context);
809 interrupt_handle_now_handler(int signal, siginfo_t *info, void *void_context)
811 os_context_t *context = arch_os_get_context(&void_context);
812 #if defined(LISP_FEATURE_LINUX) || defined(RESTORE_FP_CONTROL_FROM_CONTEXT)
813 os_restore_fp_control(context);
814 #ifndef LISP_FEATURE_WIN32
815 if ((signal == SIGILL) || (signal == SIGBUS)
816 #ifndef LISP_FEATURE_LINUX
817 || (signal == SIGEMT)
820 corruption_warning_and_maybe_lose("Signal %d recieved", signal);
823 interrupt_handle_now(signal, info, context);
826 /* manipulate the signal context and stack such that when the handler
827 * returns, it will call function instead of whatever it was doing
831 #if (defined(LISP_FEATURE_X86) || defined(LISP_FEATURE_X86_64))
832 extern int *context_eflags_addr(os_context_t *context);
835 extern lispobj call_into_lisp(lispobj fun, lispobj *args, int nargs);
836 extern void post_signal_tramp(void);
837 extern void call_into_lisp_tramp(void);
839 arrange_return_to_lisp_function(os_context_t *context, lispobj function)
841 #if !(defined(LISP_FEATURE_X86) || defined(LISP_FEATURE_X86_64))
842 void * fun=native_pointer(function);
843 void *code = &(((struct simple_fun *) fun)->code);
846 /* Build a stack frame showing `interrupted' so that the
847 * user's backtrace makes (as much) sense (as usual) */
849 /* FIXME: what about restoring fp state? */
850 /* FIXME: what about restoring errno? */
851 #ifdef LISP_FEATURE_X86
852 /* Suppose the existence of some function that saved all
853 * registers, called call_into_lisp, then restored GP registers and
854 * returned. It would look something like this:
862 pushl {address of function to call}
863 call 0x8058db0 <call_into_lisp>
870 * What we do here is set up the stack that call_into_lisp would
871 * expect to see if it had been called by this code, and frob the
872 * signal context so that signal return goes directly to call_into_lisp,
873 * and when that function (and the lisp function it invoked) returns,
874 * it returns to the second half of this imaginary function which
875 * restores all registers and returns to C
877 * For this to work, the latter part of the imaginary function
878 * must obviously exist in reality. That would be post_signal_tramp
881 u32 *sp=(u32 *)*os_context_register_addr(context,reg_ESP);
883 #if defined(LISP_FEATURE_DARWIN)
884 u32 *register_save_area = (u32 *)os_validate(0, 0x40);
886 FSHOW_SIGNAL((stderr, "/arrange_return_to_lisp_function: preparing to go to function %x, sp: %x\n", function, sp));
887 FSHOW_SIGNAL((stderr, "/arrange_return_to_lisp_function: context: %x, &context %x\n", context, &context));
889 /* 1. os_validate (malloc/mmap) register_save_block
890 * 2. copy register state into register_save_block
891 * 3. put a pointer to register_save_block in a register in the context
892 * 4. set the context's EIP to point to a trampoline which:
893 * a. builds the fake stack frame from the block
895 * c. calls the function
898 *register_save_area = *os_context_pc_addr(context);
899 *(register_save_area + 1) = function;
900 *(register_save_area + 2) = *os_context_register_addr(context,reg_EDI);
901 *(register_save_area + 3) = *os_context_register_addr(context,reg_ESI);
902 *(register_save_area + 4) = *os_context_register_addr(context,reg_EDX);
903 *(register_save_area + 5) = *os_context_register_addr(context,reg_ECX);
904 *(register_save_area + 6) = *os_context_register_addr(context,reg_EBX);
905 *(register_save_area + 7) = *os_context_register_addr(context,reg_EAX);
906 *(register_save_area + 8) = *context_eflags_addr(context);
908 *os_context_pc_addr(context) =
909 (os_context_register_t) call_into_lisp_tramp;
910 *os_context_register_addr(context,reg_ECX) =
911 (os_context_register_t) register_save_area;
914 /* return address for call_into_lisp: */
915 *(sp-15) = (u32)post_signal_tramp;
916 *(sp-14) = function; /* args for call_into_lisp : function*/
917 *(sp-13) = 0; /* arg array */
918 *(sp-12) = 0; /* no. args */
919 /* this order matches that used in POPAD */
920 *(sp-11)=*os_context_register_addr(context,reg_EDI);
921 *(sp-10)=*os_context_register_addr(context,reg_ESI);
923 *(sp-9)=*os_context_register_addr(context,reg_ESP)-8;
924 /* POPAD ignores the value of ESP: */
926 *(sp-7)=*os_context_register_addr(context,reg_EBX);
928 *(sp-6)=*os_context_register_addr(context,reg_EDX);
929 *(sp-5)=*os_context_register_addr(context,reg_ECX);
930 *(sp-4)=*os_context_register_addr(context,reg_EAX);
931 *(sp-3)=*context_eflags_addr(context);
932 *(sp-2)=*os_context_register_addr(context,reg_EBP);
933 *(sp-1)=*os_context_pc_addr(context);
937 #elif defined(LISP_FEATURE_X86_64)
938 u64 *sp=(u64 *)*os_context_register_addr(context,reg_RSP);
940 /* return address for call_into_lisp: */
941 *(sp-18) = (u64)post_signal_tramp;
943 *(sp-17)=*os_context_register_addr(context,reg_R15);
944 *(sp-16)=*os_context_register_addr(context,reg_R14);
945 *(sp-15)=*os_context_register_addr(context,reg_R13);
946 *(sp-14)=*os_context_register_addr(context,reg_R12);
947 *(sp-13)=*os_context_register_addr(context,reg_R11);
948 *(sp-12)=*os_context_register_addr(context,reg_R10);
949 *(sp-11)=*os_context_register_addr(context,reg_R9);
950 *(sp-10)=*os_context_register_addr(context,reg_R8);
951 *(sp-9)=*os_context_register_addr(context,reg_RDI);
952 *(sp-8)=*os_context_register_addr(context,reg_RSI);
953 /* skip RBP and RSP */
954 *(sp-7)=*os_context_register_addr(context,reg_RBX);
955 *(sp-6)=*os_context_register_addr(context,reg_RDX);
956 *(sp-5)=*os_context_register_addr(context,reg_RCX);
957 *(sp-4)=*os_context_register_addr(context,reg_RAX);
958 *(sp-3)=*context_eflags_addr(context);
959 *(sp-2)=*os_context_register_addr(context,reg_RBP);
960 *(sp-1)=*os_context_pc_addr(context);
962 *os_context_register_addr(context,reg_RDI) =
963 (os_context_register_t)function; /* function */
964 *os_context_register_addr(context,reg_RSI) = 0; /* arg. array */
965 *os_context_register_addr(context,reg_RDX) = 0; /* no. args */
967 struct thread *th=arch_os_get_current_thread();
968 build_fake_control_stack_frames(th,context);
971 #ifdef LISP_FEATURE_X86
973 #if !defined(LISP_FEATURE_DARWIN)
974 *os_context_pc_addr(context) = (os_context_register_t)call_into_lisp;
975 *os_context_register_addr(context,reg_ECX) = 0;
976 *os_context_register_addr(context,reg_EBP) = (os_context_register_t)(sp-2);
978 *os_context_register_addr(context,reg_UESP) =
979 (os_context_register_t)(sp-15);
981 *os_context_register_addr(context,reg_ESP) = (os_context_register_t)(sp-15);
982 #endif /* __NETBSD__ */
983 #endif /* LISP_FEATURE_DARWIN */
985 #elif defined(LISP_FEATURE_X86_64)
986 *os_context_pc_addr(context) = (os_context_register_t)call_into_lisp;
987 *os_context_register_addr(context,reg_RCX) = 0;
988 *os_context_register_addr(context,reg_RBP) = (os_context_register_t)(sp-2);
989 *os_context_register_addr(context,reg_RSP) = (os_context_register_t)(sp-18);
991 /* this much of the calling convention is common to all
993 *os_context_pc_addr(context) = (os_context_register_t)(unsigned long)code;
994 *os_context_register_addr(context,reg_NARGS) = 0;
995 *os_context_register_addr(context,reg_LIP) =
996 (os_context_register_t)(unsigned long)code;
997 *os_context_register_addr(context,reg_CFP) =
998 (os_context_register_t)(unsigned long)current_control_frame_pointer;
1000 #ifdef ARCH_HAS_NPC_REGISTER
1001 *os_context_npc_addr(context) =
1002 4 + *os_context_pc_addr(context);
1004 #ifdef LISP_FEATURE_SPARC
1005 *os_context_register_addr(context,reg_CODE) =
1006 (os_context_register_t)(fun + FUN_POINTER_LOWTAG);
1008 FSHOW((stderr, "/arranged return to Lisp function (0x%lx)\n",
1012 #ifdef LISP_FEATURE_SB_THREAD
1014 /* FIXME: this function can go away when all lisp handlers are invoked
1015 * via arrange_return_to_lisp_function. */
1017 interrupt_thread_handler(int num, siginfo_t *info, void *v_context)
1019 os_context_t *context = (os_context_t*)arch_os_get_context(&v_context);
1021 FSHOW_SIGNAL((stderr,"/interrupt_thread_handler\n"));
1022 check_blockables_blocked_or_lose();
1024 /* let the handler enable interrupts again when it sees fit */
1025 sigaddset_deferrable(os_context_sigmask_addr(context));
1026 arrange_return_to_lisp_function(context,
1027 StaticSymbolFunction(RUN_INTERRUPTION));
1032 /* KLUDGE: Theoretically the approach we use for undefined alien
1033 * variables should work for functions as well, but on PPC/Darwin
1034 * we get bus error at bogus addresses instead, hence this workaround,
1035 * that has the added benefit of automatically discriminating between
1036 * functions and variables.
1039 undefined_alien_function(void)
1041 funcall0(StaticSymbolFunction(UNDEFINED_ALIEN_FUNCTION_ERROR));
1045 handle_guard_page_triggered(os_context_t *context,os_vm_address_t addr)
1047 struct thread *th=arch_os_get_current_thread();
1049 /* note the os_context hackery here. When the signal handler returns,
1050 * it won't go back to what it was doing ... */
1051 if(addr >= CONTROL_STACK_GUARD_PAGE(th) &&
1052 addr < CONTROL_STACK_GUARD_PAGE(th) + os_vm_page_size) {
1053 /* We hit the end of the control stack: disable guard page
1054 * protection so the error handler has some headroom, protect the
1055 * previous page so that we can catch returns from the guard page
1056 * and restore it. */
1057 corruption_warning_and_maybe_lose("Control stack exhausted");
1058 protect_control_stack_guard_page(0);
1059 protect_control_stack_return_guard_page(1);
1061 arrange_return_to_lisp_function
1062 (context, StaticSymbolFunction(CONTROL_STACK_EXHAUSTED_ERROR));
1065 else if(addr >= CONTROL_STACK_RETURN_GUARD_PAGE(th) &&
1066 addr < CONTROL_STACK_RETURN_GUARD_PAGE(th) + os_vm_page_size) {
1067 /* We're returning from the guard page: reprotect it, and
1068 * unprotect this one. This works even if we somehow missed
1069 * the return-guard-page, and hit it on our way to new
1070 * exhaustion instead. */
1071 protect_control_stack_guard_page(1);
1072 protect_control_stack_return_guard_page(0);
1075 else if (addr >= undefined_alien_address &&
1076 addr < undefined_alien_address + os_vm_page_size) {
1077 arrange_return_to_lisp_function
1078 (context, StaticSymbolFunction(UNDEFINED_ALIEN_VARIABLE_ERROR));
1085 * noise to install handlers
1088 #ifndef LISP_FEATURE_WIN32
1089 /* In Linux 2.4 synchronous signals (sigtrap & co) can be delivered if
1090 * they are blocked, in Linux 2.6 the default handler is invoked
1091 * instead that usually coredumps. One might hastily think that adding
1092 * SA_NODEFER helps, but until ~2.6.13 if SA_NODEFER is specified then
1093 * the whole sa_mask is ignored and instead of not adding the signal
1094 * in question to the mask. That means if it's not blockable the
1095 * signal must be unblocked at the beginning of signal handlers.
1097 * It turns out that NetBSD's SA_NODEFER doesn't DTRT in a different
1098 * way: if SA_NODEFER is set and the signal is in sa_mask, the signal
1099 * will be unblocked in the sigmask during the signal handler. -- RMK
1102 static volatile int sigaction_nodefer_works = -1;
1104 #define SA_NODEFER_TEST_BLOCK_SIGNAL SIGABRT
1105 #define SA_NODEFER_TEST_KILL_SIGNAL SIGUSR1
1108 sigaction_nodefer_test_handler(int signal, siginfo_t *info, void *void_context)
1110 sigset_t empty, current;
1112 sigemptyset(&empty);
1113 thread_sigmask(SIG_BLOCK, &empty, ¤t);
1114 /* There should be exactly two blocked signals: the two we added
1115 * to sa_mask when setting up the handler. NetBSD doesn't block
1116 * the signal we're handling when SA_NODEFER is set; Linux before
1117 * 2.6.13 or so also doesn't block the other signal when
1118 * SA_NODEFER is set. */
1119 for(i = 1; i < NSIG; i++)
1120 if (sigismember(¤t, i) !=
1121 (((i == SA_NODEFER_TEST_BLOCK_SIGNAL) || (i == signal)) ? 1 : 0)) {
1122 FSHOW_SIGNAL((stderr, "SA_NODEFER doesn't work, signal %d\n", i));
1123 sigaction_nodefer_works = 0;
1125 if (sigaction_nodefer_works == -1)
1126 sigaction_nodefer_works = 1;
1130 see_if_sigaction_nodefer_works(void)
1132 struct sigaction sa, old_sa;
1134 sa.sa_flags = SA_SIGINFO | SA_NODEFER;
1135 sa.sa_sigaction = sigaction_nodefer_test_handler;
1136 sigemptyset(&sa.sa_mask);
1137 sigaddset(&sa.sa_mask, SA_NODEFER_TEST_BLOCK_SIGNAL);
1138 sigaddset(&sa.sa_mask, SA_NODEFER_TEST_KILL_SIGNAL);
1139 sigaction(SA_NODEFER_TEST_KILL_SIGNAL, &sa, &old_sa);
1140 /* Make sure no signals are blocked. */
1143 sigemptyset(&empty);
1144 thread_sigmask(SIG_SETMASK, &empty, 0);
1146 kill(getpid(), SA_NODEFER_TEST_KILL_SIGNAL);
1147 while (sigaction_nodefer_works == -1);
1148 sigaction(SA_NODEFER_TEST_KILL_SIGNAL, &old_sa, NULL);
1151 #undef SA_NODEFER_TEST_BLOCK_SIGNAL
1152 #undef SA_NODEFER_TEST_KILL_SIGNAL
1155 unblock_me_trampoline(int signal, siginfo_t *info, void *void_context)
1159 sigemptyset(&unblock);
1160 sigaddset(&unblock, signal);
1161 thread_sigmask(SIG_UNBLOCK, &unblock, 0);
1162 interrupt_handle_now_handler(signal, info, void_context);
1166 low_level_unblock_me_trampoline(int signal, siginfo_t *info, void *void_context)
1170 sigemptyset(&unblock);
1171 sigaddset(&unblock, signal);
1172 thread_sigmask(SIG_UNBLOCK, &unblock, 0);
1173 (*interrupt_low_level_handlers[signal])(signal, info, void_context);
1177 undoably_install_low_level_interrupt_handler (int signal,
1178 interrupt_handler_t handler)
1180 struct sigaction sa;
1182 if (0 > signal || signal >= NSIG) {
1183 lose("bad signal number %d\n", signal);
1186 if (ARE_SAME_HANDLER(handler, SIG_DFL))
1187 sa.sa_sigaction = handler;
1188 else if (sigismember(&deferrable_sigset,signal))
1189 sa.sa_sigaction = low_level_maybe_now_maybe_later;
1190 /* The use of a trampoline appears to break the
1191 arch_os_get_context() workaround for SPARC/Linux. For now,
1192 don't use the trampoline (and so be vulnerable to the problems
1193 that SA_NODEFER is meant to solve. */
1194 #if !(defined(LISP_FEATURE_SPARC) && defined(LISP_FEATURE_LINUX))
1195 else if (!sigaction_nodefer_works &&
1196 !sigismember(&blockable_sigset, signal))
1197 sa.sa_sigaction = low_level_unblock_me_trampoline;
1200 sa.sa_sigaction = handler;
1202 sigcopyset(&sa.sa_mask, &blockable_sigset);
1203 sa.sa_flags = SA_SIGINFO | SA_RESTART
1204 | (sigaction_nodefer_works ? SA_NODEFER : 0);
1205 #ifdef LISP_FEATURE_C_STACK_IS_CONTROL_STACK
1206 if((signal==SIG_MEMORY_FAULT)
1207 #ifdef SIG_INTERRUPT_THREAD
1208 || (signal==SIG_INTERRUPT_THREAD)
1211 sa.sa_flags |= SA_ONSTACK;
1214 sigaction(signal, &sa, NULL);
1215 interrupt_low_level_handlers[signal] =
1216 (ARE_SAME_HANDLER(handler, SIG_DFL) ? 0 : handler);
1220 /* This is called from Lisp. */
1222 install_handler(int signal, void handler(int, siginfo_t*, void*))
1224 #ifndef LISP_FEATURE_WIN32
1225 struct sigaction sa;
1227 union interrupt_handler oldhandler;
1229 FSHOW((stderr, "/entering POSIX install_handler(%d, ..)\n", signal));
1232 sigaddset(&new, signal);
1233 thread_sigmask(SIG_BLOCK, &new, &old);
1235 FSHOW((stderr, "/interrupt_low_level_handlers[signal]=%x\n",
1236 (unsigned int)interrupt_low_level_handlers[signal]));
1237 if (interrupt_low_level_handlers[signal]==0) {
1238 if (ARE_SAME_HANDLER(handler, SIG_DFL) ||
1239 ARE_SAME_HANDLER(handler, SIG_IGN))
1240 sa.sa_sigaction = handler;
1241 else if (sigismember(&deferrable_sigset, signal))
1242 sa.sa_sigaction = maybe_now_maybe_later;
1243 else if (!sigaction_nodefer_works &&
1244 !sigismember(&blockable_sigset, signal))
1245 sa.sa_sigaction = unblock_me_trampoline;
1247 sa.sa_sigaction = interrupt_handle_now_handler;
1249 sigcopyset(&sa.sa_mask, &blockable_sigset);
1250 sa.sa_flags = SA_SIGINFO | SA_RESTART |
1251 (sigaction_nodefer_works ? SA_NODEFER : 0);
1252 sigaction(signal, &sa, NULL);
1255 oldhandler = interrupt_handlers[signal];
1256 interrupt_handlers[signal].c = handler;
1258 thread_sigmask(SIG_SETMASK, &old, 0);
1260 FSHOW((stderr, "/leaving POSIX install_handler(%d, ..)\n", signal));
1262 return (unsigned long)oldhandler.lisp;
1264 /* Probably-wrong Win32 hack */
1269 /* This must not go through lisp as it's allowed anytime, even when on
1272 sigabrt_handler(int signal, siginfo_t *info, void *void_context)
1274 lose("SIGABRT received.\n");
1278 interrupt_init(void)
1280 #ifndef LISP_FEATURE_WIN32
1282 SHOW("entering interrupt_init()");
1283 see_if_sigaction_nodefer_works();
1284 sigemptyset(&deferrable_sigset);
1285 sigemptyset(&blockable_sigset);
1286 sigaddset_deferrable(&deferrable_sigset);
1287 sigaddset_blockable(&blockable_sigset);
1289 /* Set up high level handler information. */
1290 for (i = 0; i < NSIG; i++) {
1291 interrupt_handlers[i].c =
1292 /* (The cast here blasts away the distinction between
1293 * SA_SIGACTION-style three-argument handlers and
1294 * signal(..)-style one-argument handlers, which is OK
1295 * because it works to call the 1-argument form where the
1296 * 3-argument form is expected.) */
1297 (void (*)(int, siginfo_t*, void*))SIG_DFL;
1299 undoably_install_low_level_interrupt_handler(SIGABRT, sigabrt_handler);
1300 SHOW("returning from interrupt_init()");
1304 #ifndef LISP_FEATURE_WIN32
1306 siginfo_code(siginfo_t *info)
1308 return info->si_code;
1310 os_vm_address_t current_memory_fault_address;
1313 lisp_memory_fault_error(os_context_t *context, os_vm_address_t addr)
1315 /* FIXME: This is lossy: if we get another memory fault (eg. from
1316 * another thread) before lisp has read this, we lose the information.
1317 * However, since this is mostly informative, we'll live with that for
1318 * now -- some address is better then no address in this case.
1320 current_memory_fault_address = addr;
1321 /* To allow debugging memory faults in signal handlers and such. */
1322 corruption_warning_and_maybe_lose("Memory fault");
1323 arrange_return_to_lisp_function(context,
1324 StaticSymbolFunction(MEMORY_FAULT_ERROR));
1329 unhandled_trap_error(os_context_t *context)
1331 lispobj context_sap;
1332 fake_foreign_function_call(context);
1333 context_sap = alloc_sap(context);
1334 #ifndef LISP_FEATURE_WIN32
1335 thread_sigmask(SIG_SETMASK, os_context_sigmask_addr(context), 0);
1337 funcall1(StaticSymbolFunction(UNHANDLED_TRAP_ERROR), context_sap);
1338 lose("UNHANDLED-TRAP-ERROR fell through");
1341 /* Common logic for trapping instructions. How we actually handle each
1342 * case is highly architecture dependent, but the overall shape is
1345 handle_trap(os_context_t *context, int trap)
1348 case trap_PendingInterrupt:
1349 FSHOW((stderr, "/<trap pending interrupt>\n"));
1350 arch_skip_instruction(context);
1351 interrupt_handle_pending(context);
1355 FSHOW((stderr, "/<trap error/cerror %d>\n", trap));
1356 interrupt_internal_error(context, trap==trap_Cerror);
1358 case trap_Breakpoint:
1359 arch_handle_breakpoint(context);
1361 case trap_FunEndBreakpoint:
1362 arch_handle_fun_end_breakpoint(context);
1364 #ifdef trap_AfterBreakpoint
1365 case trap_AfterBreakpoint:
1366 arch_handle_after_breakpoint(context);
1369 #ifdef trap_SingleStepAround
1370 case trap_SingleStepAround:
1371 case trap_SingleStepBefore:
1372 arch_handle_single_step_trap(context, trap);
1376 fake_foreign_function_call(context);
1377 lose("%%PRIMITIVE HALT called; the party is over.\n");
1379 unhandled_trap_error(context);