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 "pseudo-atomic.h"
67 #include "genesis/fdefn.h"
68 #include "genesis/simple-fun.h"
69 #include "genesis/cons.h"
71 /* When we catch an internal error, should we pass it back to Lisp to
72 * be handled in a high-level way? (Early in cold init, the answer is
73 * 'no', because Lisp is still too brain-dead to handle anything.
74 * After sufficient initialization has been completed, the answer
76 boolean internal_errors_enabled = 0;
78 #ifndef LISP_FEATURE_WIN32
80 void (*interrupt_low_level_handlers[NSIG]) (int, siginfo_t*, os_context_t*);
82 union interrupt_handler interrupt_handlers[NSIG];
84 /* Under Linux on some architectures, we appear to have to restore the
85 * FPU control word from the context, as after the signal is delivered
86 * we appear to have a null FPU control word. */
87 #if defined(RESTORE_FP_CONTROL_FROM_CONTEXT)
88 #define RESTORE_FP_CONTROL_WORD(context,void_context) \
89 os_context_t *context = arch_os_get_context(&void_context); \
90 os_restore_fp_control(context);
92 #define RESTORE_FP_CONTROL_WORD(context,void_context) \
93 os_context_t *context = arch_os_get_context(&void_context);
96 /* Foreign code may want to start some threads on its own.
97 * Non-targetted, truly asynchronous signals can be delivered to
98 * basically any thread, but invoking Lisp handlers in such foregign
99 * threads is really bad, so let's resignal it.
101 * This should at least bring attention to the problem, but it cannot
102 * work for SIGSEGV and similar. It is good enough for timers, and
103 * maybe all deferrables. */
105 #ifdef LISP_FEATURE_SB_THREAD
107 add_handled_signals(sigset_t *sigset)
110 for(i = 1; i < NSIG; i++) {
111 if (!(ARE_SAME_HANDLER(interrupt_low_level_handlers[i], SIG_DFL)) ||
112 !(ARE_SAME_HANDLER(interrupt_handlers[i].c, SIG_DFL))) {
113 sigaddset(sigset, i);
118 void block_signals(sigset_t *what, sigset_t *where, sigset_t *old);
122 maybe_resignal_to_lisp_thread(int signal, os_context_t *context)
124 #ifdef LISP_FEATURE_SB_THREAD
125 if (!pthread_getspecific(lisp_thread)) {
126 if (!(sigismember(&deferrable_sigset,signal))) {
127 corruption_warning_and_maybe_lose
128 ("Received signal %d in non-lisp thread %lu, resignalling to a lisp thread.",
134 sigemptyset(&sigset);
135 add_handled_signals(&sigset);
136 block_signals(&sigset, 0, 0);
137 block_signals(&sigset, os_context_sigmask_addr(context), 0);
138 kill(getpid(), signal);
146 /* These are to be used in signal handlers. Currently all handlers are
147 * called from one of:
149 * interrupt_handle_now_handler
150 * maybe_now_maybe_later
151 * unblock_me_trampoline
152 * low_level_handle_now_handler
153 * low_level_maybe_now_maybe_later
154 * low_level_unblock_me_trampoline
156 * This gives us a single point of control (or six) over errno, fp
157 * control word, and fixing up signal context on sparc.
159 * The SPARC/Linux platform doesn't quite do signals the way we want
160 * them done. The third argument in the handler isn't filled in by the
161 * kernel properly, so we fix it up ourselves in the
162 * arch_os_get_context(..) function. -- CSR, 2002-07-23
164 #define SAVE_ERRNO(signal,context,void_context) \
166 int _saved_errno = errno; \
167 RESTORE_FP_CONTROL_WORD(context,void_context); \
168 if (!maybe_resignal_to_lisp_thread(signal, context)) \
171 #define RESTORE_ERRNO \
173 errno = _saved_errno; \
176 static void run_deferred_handler(struct interrupt_data *data,
177 os_context_t *context);
178 #ifndef LISP_FEATURE_WIN32
179 static void store_signal_data_for_later (struct interrupt_data *data,
180 void *handler, int signal,
182 os_context_t *context);
185 /* Generic signal related utilities. */
188 get_current_sigmask(sigset_t *sigset)
190 /* Get the current sigmask, by blocking the empty set. */
191 thread_sigmask(SIG_BLOCK, 0, sigset);
195 block_signals(sigset_t *what, sigset_t *where, sigset_t *old)
200 sigcopyset(old, where);
201 for(i = 1; i < NSIG; i++) {
202 if (sigismember(what, i))
206 thread_sigmask(SIG_BLOCK, what, old);
211 unblock_signals(sigset_t *what, sigset_t *where, sigset_t *old)
216 sigcopyset(old, where);
217 for(i = 1; i < NSIG; i++) {
218 if (sigismember(what, i))
222 thread_sigmask(SIG_UNBLOCK, what, old);
227 print_sigset(sigset_t *sigset)
230 for(i = 1; i < NSIG; i++) {
231 if (sigismember(sigset, i))
232 fprintf(stderr, "Signal %d masked\n", i);
236 /* Return 1 is all signals is sigset2 are masked in sigset, return 0
237 * if all re unmasked else die. Passing NULL for sigset is a shorthand
238 * for the current sigmask. */
240 all_signals_blocked_p(sigset_t *sigset, sigset_t *sigset2,
243 #if !defined(LISP_FEATURE_WIN32)
245 boolean has_blocked = 0, has_unblocked = 0;
248 get_current_sigmask(¤t);
251 for(i = 1; i < NSIG; i++) {
252 if (sigismember(sigset2, i)) {
253 if (sigismember(sigset, i))
259 if (has_blocked && has_unblocked) {
260 print_sigset(sigset);
261 lose("some %s signals blocked, some unblocked\n", name);
271 /* Deferrables, blockables, gc signals. */
274 sigaddset_deferrable(sigset_t *s)
276 sigaddset(s, SIGHUP);
277 sigaddset(s, SIGINT);
278 sigaddset(s, SIGTERM);
279 sigaddset(s, SIGQUIT);
280 sigaddset(s, SIGPIPE);
281 sigaddset(s, SIGALRM);
282 sigaddset(s, SIGURG);
283 sigaddset(s, SIGTSTP);
284 sigaddset(s, SIGCHLD);
286 #ifndef LISP_FEATURE_HPUX
287 sigaddset(s, SIGXCPU);
288 sigaddset(s, SIGXFSZ);
290 sigaddset(s, SIGVTALRM);
291 sigaddset(s, SIGPROF);
292 sigaddset(s, SIGWINCH);
296 sigaddset_blockable(sigset_t *sigset)
298 sigaddset_deferrable(sigset);
299 sigaddset_gc(sigset);
303 sigaddset_gc(sigset_t *sigset)
305 #ifdef LISP_FEATURE_SB_THREAD
306 sigaddset(sigset,SIG_STOP_FOR_GC);
310 /* initialized in interrupt_init */
311 sigset_t deferrable_sigset;
312 sigset_t blockable_sigset;
317 #if !defined(LISP_FEATURE_WIN32)
319 deferrables_blocked_p(sigset_t *sigset)
321 return all_signals_blocked_p(sigset, &deferrable_sigset, "deferrable");
326 check_deferrables_unblocked_or_lose(sigset_t *sigset)
328 #if !defined(LISP_FEATURE_WIN32)
329 if (deferrables_blocked_p(sigset))
330 lose("deferrables blocked\n");
335 check_deferrables_blocked_or_lose(sigset_t *sigset)
337 #if !defined(LISP_FEATURE_WIN32)
338 if (!deferrables_blocked_p(sigset))
339 lose("deferrables unblocked\n");
343 #if !defined(LISP_FEATURE_WIN32)
345 blockables_blocked_p(sigset_t *sigset)
347 return all_signals_blocked_p(sigset, &blockable_sigset, "blockable");
352 check_blockables_unblocked_or_lose(sigset_t *sigset)
354 #if !defined(LISP_FEATURE_WIN32)
355 if (blockables_blocked_p(sigset))
356 lose("blockables blocked\n");
361 check_blockables_blocked_or_lose(sigset_t *sigset)
363 #if !defined(LISP_FEATURE_WIN32)
364 if (!blockables_blocked_p(sigset))
365 lose("blockables unblocked\n");
369 #if !defined(LISP_FEATURE_WIN32)
371 gc_signals_blocked_p(sigset_t *sigset)
373 return all_signals_blocked_p(sigset, &gc_sigset, "gc");
378 check_gc_signals_unblocked_or_lose(sigset_t *sigset)
380 #if !defined(LISP_FEATURE_WIN32)
381 if (gc_signals_blocked_p(sigset))
382 lose("gc signals blocked\n");
387 check_gc_signals_blocked_or_lose(sigset_t *sigset)
389 #if !defined(LISP_FEATURE_WIN32)
390 if (!gc_signals_blocked_p(sigset))
391 lose("gc signals unblocked\n");
396 block_deferrable_signals(sigset_t *where, sigset_t *old)
398 #ifndef LISP_FEATURE_WIN32
399 block_signals(&deferrable_sigset, where, old);
404 block_blockable_signals(sigset_t *where, sigset_t *old)
406 #ifndef LISP_FEATURE_WIN32
407 block_signals(&blockable_sigset, where, old);
412 block_gc_signals(sigset_t *where, sigset_t *old)
414 #ifndef LISP_FEATURE_WIN32
415 block_signals(&gc_sigset, where, old);
420 unblock_deferrable_signals(sigset_t *where, sigset_t *old)
422 #ifndef LISP_FEATURE_WIN32
423 if (interrupt_handler_pending_p())
424 lose("unblock_deferrable_signals: losing proposition\n");
425 check_gc_signals_unblocked_or_lose(where);
426 unblock_signals(&deferrable_sigset, where, old);
431 unblock_blockable_signals(sigset_t *where, sigset_t *old)
433 #ifndef LISP_FEATURE_WIN32
434 unblock_signals(&blockable_sigset, where, old);
439 unblock_gc_signals(sigset_t *where, sigset_t *old)
441 #ifndef LISP_FEATURE_WIN32
442 unblock_signals(&gc_sigset, where, old);
447 unblock_signals_in_context_and_maybe_warn(os_context_t *context)
449 #ifndef LISP_FEATURE_WIN32
450 sigset_t *sigset = os_context_sigmask_addr(context);
451 if (all_signals_blocked_p(sigset, &gc_sigset, "gc")) {
452 corruption_warning_and_maybe_lose(
453 "Enabling blocked gc signals to allow returning to Lisp without risking\n\
454 gc deadlocks. Since GC signals are only blocked in signal handlers when \n\
455 they are not safe to interrupt at all, this is a pretty severe occurrence.\n");
456 unblock_gc_signals(sigset, 0);
458 if (!interrupt_handler_pending_p()) {
459 unblock_deferrable_signals(sigset, 0);
466 check_interrupts_enabled_or_lose(os_context_t *context)
468 struct thread *thread=arch_os_get_current_thread();
469 if (SymbolValue(INTERRUPTS_ENABLED,thread) == NIL)
470 lose("interrupts not enabled\n");
471 if (arch_pseudo_atomic_atomic(context))
472 lose ("in pseudo atomic section\n");
475 /* Save sigset (or the current sigmask if 0) if there is no pending
476 * handler, because that means that deferabbles are already blocked.
477 * The purpose is to avoid losing the pending gc signal if a
478 * deferrable interrupt async unwinds between clearing the pseudo
479 * atomic and trapping to GC.*/
481 maybe_save_gc_mask_and_block_deferrables(sigset_t *sigset)
483 #ifndef LISP_FEATURE_WIN32
484 struct thread *thread = arch_os_get_current_thread();
485 struct interrupt_data *data = thread->interrupt_data;
487 /* Obviously, this function is called when signals may not be
488 * blocked. Let's make sure we are not interrupted. */
489 block_blockable_signals(0, &oldset);
490 #ifndef LISP_FEATURE_SB_THREAD
491 /* With threads a SIG_STOP_FOR_GC and a normal GC may also want to
493 if (data->gc_blocked_deferrables)
494 lose("gc_blocked_deferrables already true\n");
496 if ((!data->pending_handler) &&
497 (!data->gc_blocked_deferrables)) {
498 FSHOW_SIGNAL((stderr,"/setting gc_blocked_deferrables\n"));
499 data->gc_blocked_deferrables = 1;
501 /* This is the sigmask of some context. */
502 sigcopyset(&data->pending_mask, sigset);
503 sigaddset_deferrable(sigset);
504 thread_sigmask(SIG_SETMASK,&oldset,0);
507 /* Operating on the current sigmask. Save oldset and
508 * unblock gc signals. In the end, this is equivalent to
509 * blocking the deferrables. */
510 sigcopyset(&data->pending_mask, &oldset);
511 thread_sigmask(SIG_UNBLOCK, &gc_sigset, 0);
515 thread_sigmask(SIG_SETMASK,&oldset,0);
519 /* Are we leaving WITH-GCING and already running with interrupts
520 * enabled, without the protection of *GC-INHIBIT* T and there is gc
521 * (or stop for gc) pending, but we haven't trapped yet? */
523 in_leaving_without_gcing_race_p(struct thread *thread)
525 return ((SymbolValue(IN_WITHOUT_GCING,thread) != NIL) &&
526 (SymbolValue(INTERRUPTS_ENABLED,thread) != NIL) &&
527 (SymbolValue(GC_INHIBIT,thread) == NIL) &&
528 ((SymbolValue(GC_PENDING,thread) != NIL)
529 #if defined(LISP_FEATURE_SB_THREAD)
530 || (SymbolValue(STOP_FOR_GC_PENDING,thread) != NIL)
535 /* Check our baroque invariants. */
537 check_interrupt_context_or_lose(os_context_t *context)
539 #ifndef LISP_FEATURE_WIN32
540 struct thread *thread = arch_os_get_current_thread();
541 struct interrupt_data *data = thread->interrupt_data;
542 int interrupt_deferred_p = (data->pending_handler != 0);
543 int interrupt_pending = (SymbolValue(INTERRUPT_PENDING,thread) != NIL);
544 sigset_t *sigset = os_context_sigmask_addr(context);
545 /* On PPC pseudo_atomic_interrupted is cleared when coming out of
546 * handle_allocation_trap. */
547 #if defined(LISP_FEATURE_GENCGC) && !defined(LISP_FEATURE_PPC)
548 int interrupts_enabled = (SymbolValue(INTERRUPTS_ENABLED,thread) != NIL);
549 int gc_inhibit = (SymbolValue(GC_INHIBIT,thread) != NIL);
550 int gc_pending = (SymbolValue(GC_PENDING,thread) == T);
551 int pseudo_atomic_interrupted = get_pseudo_atomic_interrupted(thread);
552 int in_race_p = in_leaving_without_gcing_race_p(thread);
553 /* In the time window between leaving the *INTERRUPTS-ENABLED* NIL
554 * section and trapping, a SIG_STOP_FOR_GC would see the next
555 * check fail, for this reason sig_stop_for_gc handler does not
556 * call this function. */
557 if (interrupt_deferred_p) {
558 if (!(!interrupts_enabled || pseudo_atomic_interrupted || in_race_p))
559 lose("Stray deferred interrupt.\n");
562 if (!(pseudo_atomic_interrupted || gc_inhibit || in_race_p))
563 lose("GC_PENDING, but why?\n");
564 #if defined(LISP_FEATURE_SB_THREAD)
566 int stop_for_gc_pending =
567 (SymbolValue(STOP_FOR_GC_PENDING,thread) != NIL);
568 if (stop_for_gc_pending)
569 if (!(pseudo_atomic_interrupted || gc_inhibit || in_race_p))
570 lose("STOP_FOR_GC_PENDING, but why?\n");
571 if (pseudo_atomic_interrupted)
572 if (!(gc_pending || stop_for_gc_pending || interrupt_deferred_p))
573 lose("pseudo_atomic_interrupted, but why?\n");
576 if (pseudo_atomic_interrupted)
577 if (!(gc_pending || interrupt_deferred_p))
578 lose("pseudo_atomic_interrupted, but why?\n");
581 if (interrupt_pending && !interrupt_deferred_p)
582 lose("INTERRUPT_PENDING but not pending handler.\n");
583 if ((data->gc_blocked_deferrables) && interrupt_pending)
584 lose("gc_blocked_deferrables and interrupt pending\n.");
585 if (data->gc_blocked_deferrables)
586 check_deferrables_blocked_or_lose(sigset);
587 if (interrupt_pending || interrupt_deferred_p ||
588 data->gc_blocked_deferrables)
589 check_deferrables_blocked_or_lose(sigset);
591 check_deferrables_unblocked_or_lose(sigset);
592 /* If deferrables are unblocked then we are open to signals
593 * that run lisp code. */
594 check_gc_signals_unblocked_or_lose(sigset);
600 * utility routines used by various signal handlers
604 build_fake_control_stack_frames(struct thread *th,os_context_t *context)
606 #ifndef LISP_FEATURE_C_STACK_IS_CONTROL_STACK
610 /* Build a fake stack frame or frames */
612 access_control_frame_pointer(th) =
613 (lispobj *)(unsigned long)
614 (*os_context_register_addr(context, reg_CSP));
615 if ((lispobj *)(unsigned long)
616 (*os_context_register_addr(context, reg_CFP))
617 == access_control_frame_pointer(th)) {
618 /* There is a small window during call where the callee's
619 * frame isn't built yet. */
620 if (lowtag_of(*os_context_register_addr(context, reg_CODE))
621 == FUN_POINTER_LOWTAG) {
622 /* We have called, but not built the new frame, so
623 * build it for them. */
624 access_control_frame_pointer(th)[0] =
625 *os_context_register_addr(context, reg_OCFP);
626 access_control_frame_pointer(th)[1] =
627 *os_context_register_addr(context, reg_LRA);
628 access_control_frame_pointer(th) += 8;
629 /* Build our frame on top of it. */
630 oldcont = (lispobj)(*os_context_register_addr(context, reg_CFP));
633 /* We haven't yet called, build our frame as if the
634 * partial frame wasn't there. */
635 oldcont = (lispobj)(*os_context_register_addr(context, reg_OCFP));
638 /* We can't tell whether we are still in the caller if it had to
639 * allocate a stack frame due to stack arguments. */
640 /* This observation provoked some past CMUCL maintainer to ask
641 * "Can anything strange happen during return?" */
644 oldcont = (lispobj)(*os_context_register_addr(context, reg_CFP));
647 access_control_stack_pointer(th) = access_control_frame_pointer(th) + 8;
649 access_control_frame_pointer(th)[0] = oldcont;
650 access_control_frame_pointer(th)[1] = NIL;
651 access_control_frame_pointer(th)[2] =
652 (lispobj)(*os_context_register_addr(context, reg_CODE));
656 /* Stores the context for gc to scavange and builds fake stack
659 fake_foreign_function_call(os_context_t *context)
662 struct thread *thread=arch_os_get_current_thread();
664 /* context_index incrementing must not be interrupted */
665 check_blockables_blocked_or_lose(0);
667 /* Get current Lisp state from context. */
669 #ifdef LISP_FEATURE_SB_THREAD
670 thread->pseudo_atomic_bits =
672 dynamic_space_free_pointer =
673 (lispobj *)(unsigned long)
675 (*os_context_register_addr(context, reg_ALLOC));
676 /* fprintf(stderr,"dynamic_space_free_pointer: %p\n", */
677 /* dynamic_space_free_pointer); */
678 #if defined(LISP_FEATURE_ALPHA) || defined(LISP_FEATURE_MIPS)
679 if ((long)dynamic_space_free_pointer & 1) {
680 lose("dead in fake_foreign_function_call, context = %x\n", context);
683 /* why doesnt PPC and SPARC do something like this: */
684 #if defined(LISP_FEATURE_HPPA)
685 if ((long)dynamic_space_free_pointer & 4) {
686 lose("dead in fake_foreign_function_call, context = %x, d_s_f_p = %x\n", context, dynamic_space_free_pointer);
691 set_binding_stack_pointer(thread,
692 *os_context_register_addr(context, reg_BSP));
695 build_fake_control_stack_frames(thread,context);
697 /* Do dynamic binding of the active interrupt context index
698 * and save the context in the context array. */
700 fixnum_value(SymbolValue(FREE_INTERRUPT_CONTEXT_INDEX,thread));
702 if (context_index >= MAX_INTERRUPTS) {
703 lose("maximum interrupt nesting depth (%d) exceeded\n", MAX_INTERRUPTS);
706 bind_variable(FREE_INTERRUPT_CONTEXT_INDEX,
707 make_fixnum(context_index + 1),thread);
709 thread->interrupt_contexts[context_index] = context;
711 #if !defined(LISP_FEATURE_X86) && !defined(LISP_FEATURE_X86_64)
712 /* x86oid targets don't maintain the foreign function call flag at
713 * all, so leave them to believe that they are never in foreign
715 foreign_function_call_active_p(thread) = 1;
719 /* blocks all blockable signals. If you are calling from a signal handler,
720 * the usual signal mask will be restored from the context when the handler
721 * finishes. Otherwise, be careful */
723 undo_fake_foreign_function_call(os_context_t *context)
725 struct thread *thread=arch_os_get_current_thread();
726 /* Block all blockable signals. */
727 block_blockable_signals(0, 0);
729 foreign_function_call_active_p(thread) = 0;
731 /* Undo dynamic binding of FREE_INTERRUPT_CONTEXT_INDEX */
734 #if defined(reg_ALLOC) && !defined(LISP_FEATURE_SB_THREAD)
735 /* Put the dynamic space free pointer back into the context. */
736 *os_context_register_addr(context, reg_ALLOC) =
737 (unsigned long) dynamic_space_free_pointer
738 | (*os_context_register_addr(context, reg_ALLOC)
741 ((unsigned long)(*os_context_register_addr(context, reg_ALLOC))
743 | ((unsigned long) dynamic_space_free_pointer & LOWTAG_MASK);
746 #if defined(reg_ALLOC) && defined(LISP_FEATURE_SB_THREAD)
747 /* Put the pseudo-atomic bits and dynamic space free pointer back
748 * into the context (p-a-bits for p-a, and dynamic space free
749 * pointer for ROOM). */
750 *os_context_register_addr(context, reg_ALLOC) =
751 (unsigned long) dynamic_space_free_pointer
752 | (thread->pseudo_atomic_bits & LOWTAG_MASK);
753 /* And clear them so we don't get bit later by call-in/call-out
754 * not updating them. */
755 thread->pseudo_atomic_bits = 0;
759 /* a handler for the signal caused by execution of a trap opcode
760 * signalling an internal error */
762 interrupt_internal_error(os_context_t *context, boolean continuable)
766 fake_foreign_function_call(context);
768 if (!internal_errors_enabled) {
769 describe_internal_error(context);
770 /* There's no good way to recover from an internal error
771 * before the Lisp error handling mechanism is set up. */
772 lose("internal error too early in init, can't recover\n");
775 /* Allocate the SAP object while the interrupts are still
777 unblock_gc_signals(0, 0);
778 context_sap = alloc_sap(context);
780 #ifndef LISP_FEATURE_WIN32
781 thread_sigmask(SIG_SETMASK, os_context_sigmask_addr(context), 0);
784 #if defined(LISP_FEATURE_LINUX) && defined(LISP_FEATURE_MIPS)
785 /* Workaround for blocked SIGTRAP. */
788 sigemptyset(&newset);
789 sigaddset(&newset, SIGTRAP);
790 thread_sigmask(SIG_UNBLOCK, &newset, 0);
794 SHOW("in interrupt_internal_error");
796 /* Display some rudimentary debugging information about the
797 * error, so that even if the Lisp error handler gets badly
798 * confused, we have a chance to determine what's going on. */
799 describe_internal_error(context);
801 funcall2(StaticSymbolFunction(INTERNAL_ERROR), context_sap,
802 continuable ? T : NIL);
804 undo_fake_foreign_function_call(context); /* blocks signals again */
806 arch_skip_instruction(context);
810 interrupt_handler_pending_p(void)
812 struct thread *thread = arch_os_get_current_thread();
813 struct interrupt_data *data = thread->interrupt_data;
814 return (data->pending_handler != 0);
818 interrupt_handle_pending(os_context_t *context)
820 /* There are three ways we can get here. First, if an interrupt
821 * occurs within pseudo-atomic, it will be deferred, and we'll
822 * trap to here at the end of the pseudo-atomic block. Second, if
823 * the GC (in alloc()) decides that a GC is required, it will set
824 * *GC-PENDING* and pseudo-atomic-interrupted if not *GC-INHIBIT*,
825 * and alloc() is always called from within pseudo-atomic, and
826 * thus we end up here again. Third, when calling GC-ON or at the
827 * end of a WITHOUT-GCING, MAYBE-HANDLE-PENDING-GC will trap to
828 * here if there is a pending GC. Fourth, ahem, at the end of
829 * WITHOUT-INTERRUPTS (bar complications with nesting). */
831 /* Win32 only needs to handle the GC cases (for now?) */
833 struct thread *thread = arch_os_get_current_thread();
834 struct interrupt_data *data = thread->interrupt_data;
836 if (arch_pseudo_atomic_atomic(context)) {
837 lose("Handling pending interrupt in pseudo atomic.");
840 FSHOW_SIGNAL((stderr, "/entering interrupt_handle_pending\n"));
842 check_blockables_blocked_or_lose(0);
844 /* If GC/SIG_STOP_FOR_GC struck during PA and there was no pending
845 * handler, then the pending mask was saved and
846 * gc_blocked_deferrables set. Hence, there can be no pending
847 * handler and it's safe to restore the pending mask.
849 * Note, that if gc_blocked_deferrables is false we may still have
850 * to GC. In this case, we are coming out of a WITHOUT-GCING or a
851 * pseudo atomic was interrupt be a deferrable first. */
852 if (data->gc_blocked_deferrables) {
853 if (data->pending_handler)
854 lose("GC blocked deferrables but still got a pending handler.");
855 if (SymbolValue(GC_INHIBIT,thread)!=NIL)
856 lose("GC blocked deferrables while GC is inhibited.");
857 /* Restore the saved signal mask from the original signal (the
858 * one that interrupted us during the critical section) into
859 * the os_context for the signal we're currently in the
860 * handler for. This should ensure that when we return from
861 * the handler the blocked signals are unblocked. */
862 #ifndef LISP_FEATURE_WIN32
863 sigcopyset(os_context_sigmask_addr(context), &data->pending_mask);
865 data->gc_blocked_deferrables = 0;
868 if (SymbolValue(GC_INHIBIT,thread)==NIL) {
869 void *original_pending_handler = data->pending_handler;
871 #ifdef LISP_FEATURE_SB_THREAD
872 if (SymbolValue(STOP_FOR_GC_PENDING,thread) != NIL) {
873 /* STOP_FOR_GC_PENDING and GC_PENDING are cleared by
874 * the signal handler if it actually stops us. */
875 arch_clear_pseudo_atomic_interrupted(context);
876 sig_stop_for_gc_handler(SIG_STOP_FOR_GC,NULL,context);
879 /* Test for T and not for != NIL since the value :IN-PROGRESS
880 * is used in SUB-GC as part of the mechanism to supress
882 if (SymbolValue(GC_PENDING,thread) == T) {
884 /* Two reasons for doing this. First, if there is a
885 * pending handler we don't want to run. Second, we are
886 * going to clear pseudo atomic interrupted to avoid
887 * spurious trapping on every allocation in SUB_GC and
888 * having a pending handler with interrupts enabled and
889 * without pseudo atomic interrupted breaks an
891 if (data->pending_handler) {
892 bind_variable(ALLOW_WITH_INTERRUPTS, NIL, thread);
893 bind_variable(INTERRUPTS_ENABLED, NIL, thread);
896 arch_clear_pseudo_atomic_interrupted(context);
898 /* GC_PENDING is cleared in SUB-GC, or if another thread
899 * is doing a gc already we will get a SIG_STOP_FOR_GC and
900 * that will clear it.
902 * If there is a pending handler or gc was triggerred in a
903 * signal handler then maybe_gc won't run POST_GC and will
904 * return normally. */
905 if (!maybe_gc(context))
906 lose("GC not inhibited but maybe_gc did not GC.");
908 if (data->pending_handler) {
912 } else if (SymbolValue(GC_PENDING,thread) != NIL) {
913 /* It's not NIL or T so GC_PENDING is :IN-PROGRESS. If
914 * GC-PENDING is not NIL then we cannot trap on pseudo
915 * atomic due to GC (see if(GC_PENDING) logic in
916 * cheneygc.c an gengcgc.c), plus there is a outer
917 * WITHOUT-INTERRUPTS SUB_GC, so how did we end up
919 lose("Trapping to run pending handler while GC in progress.");
922 check_blockables_blocked_or_lose(0);
924 /* No GC shall be lost. If SUB_GC triggers another GC then
925 * that should be handled on the spot. */
926 if (SymbolValue(GC_PENDING,thread) != NIL)
927 lose("GC_PENDING after doing gc.");
928 #ifdef LISP_FEATURE_SB_THREAD
929 if (SymbolValue(STOP_FOR_GC_PENDING,thread) != NIL)
930 lose("STOP_FOR_GC_PENDING after doing gc.");
932 /* Check two things. First, that gc does not clobber a handler
933 * that's already pending. Second, that there is no interrupt
934 * lossage: if original_pending_handler was NULL then even if
935 * an interrupt arrived during GC (POST-GC, really) it was
937 if (original_pending_handler != data->pending_handler)
938 lose("pending handler changed in gc: %x -> %d.",
939 original_pending_handler, data->pending_handler);
942 #ifndef LISP_FEATURE_WIN32
943 /* There may be no pending handler, because it was only a gc that
944 * had to be executed or because Lisp is a bit too eager to call
945 * DO-PENDING-INTERRUPT. */
946 if ((SymbolValue(INTERRUPTS_ENABLED,thread) != NIL) &&
947 (data->pending_handler)) {
948 /* No matter how we ended up here, clear both
949 * INTERRUPT_PENDING and pseudo atomic interrupted. It's safe
950 * because we checked above that there is no GC pending. */
951 SetSymbolValue(INTERRUPT_PENDING, NIL, thread);
952 arch_clear_pseudo_atomic_interrupted(context);
953 /* Restore the sigmask in the context. */
954 sigcopyset(os_context_sigmask_addr(context), &data->pending_mask);
955 run_deferred_handler(data, context);
958 #ifdef LISP_FEATURE_GENCGC
959 if (get_pseudo_atomic_interrupted(thread))
960 lose("pseudo_atomic_interrupted after interrupt_handle_pending\n");
962 /* It is possible that the end of this function was reached
963 * without never actually doing anything, the tests in Lisp for
964 * when to call receive-pending-interrupt are not exact. */
965 FSHOW_SIGNAL((stderr, "/exiting interrupt_handle_pending\n"));
970 interrupt_handle_now(int signal, siginfo_t *info, os_context_t *context)
972 boolean were_in_lisp;
973 union interrupt_handler handler;
975 check_blockables_blocked_or_lose(0);
977 #ifndef LISP_FEATURE_WIN32
978 if (sigismember(&deferrable_sigset,signal))
979 check_interrupts_enabled_or_lose(context);
982 handler = interrupt_handlers[signal];
984 if (ARE_SAME_HANDLER(handler.c, SIG_IGN)) {
988 were_in_lisp = !foreign_function_call_active_p(arch_os_get_current_thread());
991 fake_foreign_function_call(context);
994 FSHOW_SIGNAL((stderr,
995 "/entering interrupt_handle_now(%d, info, context)\n",
998 if (ARE_SAME_HANDLER(handler.c, SIG_DFL)) {
1000 /* This can happen if someone tries to ignore or default one
1001 * of the signals we need for runtime support, and the runtime
1002 * support decides to pass on it. */
1003 lose("no handler for signal %d in interrupt_handle_now(..)\n", signal);
1005 } else if (lowtag_of(handler.lisp) == FUN_POINTER_LOWTAG) {
1006 /* Once we've decided what to do about contexts in a
1007 * return-elsewhere world (the original context will no longer
1008 * be available; should we copy it or was nobody using it anyway?)
1009 * then we should convert this to return-elsewhere */
1011 /* CMUCL comment said "Allocate the SAPs while the interrupts
1012 * are still disabled.". I (dan, 2003.08.21) assume this is
1013 * because we're not in pseudoatomic and allocation shouldn't
1014 * be interrupted. In which case it's no longer an issue as
1015 * all our allocation from C now goes through a PA wrapper,
1016 * but still, doesn't hurt.
1018 * Yeah, but non-gencgc platforms don't really wrap allocation
1019 * in PA. MG - 2005-08-29 */
1021 lispobj info_sap, context_sap;
1022 /* Leave deferrable signals blocked, the handler itself will
1023 * allow signals again when it sees fit. */
1024 unblock_gc_signals(0, 0);
1025 context_sap = alloc_sap(context);
1026 info_sap = alloc_sap(info);
1028 FSHOW_SIGNAL((stderr,"/calling Lisp-level handler\n"));
1030 funcall3(handler.lisp,
1031 make_fixnum(signal),
1035 /* This cannot happen in sane circumstances. */
1037 FSHOW_SIGNAL((stderr,"/calling C-level handler\n"));
1039 #ifndef LISP_FEATURE_WIN32
1040 /* Allow signals again. */
1041 thread_sigmask(SIG_SETMASK, os_context_sigmask_addr(context), 0);
1043 (*handler.c)(signal, info, context);
1048 undo_fake_foreign_function_call(context); /* block signals again */
1051 FSHOW_SIGNAL((stderr,
1052 "/returning from interrupt_handle_now(%d, info, context)\n",
1056 /* This is called at the end of a critical section if the indications
1057 * are that some signal was deferred during the section. Note that as
1058 * far as C or the kernel is concerned we dealt with the signal
1059 * already; we're just doing the Lisp-level processing now that we
1062 run_deferred_handler(struct interrupt_data *data, os_context_t *context)
1064 /* The pending_handler may enable interrupts and then another
1065 * interrupt may hit, overwrite interrupt_data, so reset the
1066 * pending handler before calling it. Trust the handler to finish
1067 * with the siginfo before enabling interrupts. */
1068 void (*pending_handler) (int, siginfo_t*, os_context_t*) =
1069 data->pending_handler;
1071 data->pending_handler=0;
1072 FSHOW_SIGNAL((stderr, "/running deferred handler %p\n", pending_handler));
1073 (*pending_handler)(data->pending_signal,&(data->pending_info), context);
1076 #ifndef LISP_FEATURE_WIN32
1078 maybe_defer_handler(void *handler, struct interrupt_data *data,
1079 int signal, siginfo_t *info, os_context_t *context)
1081 struct thread *thread=arch_os_get_current_thread();
1083 check_blockables_blocked_or_lose(0);
1085 if (SymbolValue(INTERRUPT_PENDING,thread) != NIL)
1086 lose("interrupt already pending\n");
1087 if (thread->interrupt_data->pending_handler)
1088 lose("there is a pending handler already (PA)\n");
1089 if (data->gc_blocked_deferrables)
1090 lose("maybe_defer_handler: gc_blocked_deferrables true\n");
1091 check_interrupt_context_or_lose(context);
1092 /* If interrupts are disabled then INTERRUPT_PENDING is set and
1093 * not PSEDUO_ATOMIC_INTERRUPTED. This is important for a pseudo
1094 * atomic section inside a WITHOUT-INTERRUPTS.
1096 * Also, if in_leaving_without_gcing_race_p then
1097 * interrupt_handle_pending is going to be called soon, so
1098 * stashing the signal away is safe.
1100 if ((SymbolValue(INTERRUPTS_ENABLED,thread) == NIL) ||
1101 in_leaving_without_gcing_race_p(thread)) {
1102 FSHOW_SIGNAL((stderr,
1103 "/maybe_defer_handler(%x,%d): deferred (RACE=%d)\n",
1104 (unsigned int)handler,signal,
1105 in_leaving_without_gcing_race_p(thread)));
1106 store_signal_data_for_later(data,handler,signal,info,context);
1107 SetSymbolValue(INTERRUPT_PENDING, T,thread);
1108 check_interrupt_context_or_lose(context);
1111 /* a slightly confusing test. arch_pseudo_atomic_atomic() doesn't
1112 * actually use its argument for anything on x86, so this branch
1113 * may succeed even when context is null (gencgc alloc()) */
1114 if (arch_pseudo_atomic_atomic(context)) {
1115 FSHOW_SIGNAL((stderr,
1116 "/maybe_defer_handler(%x,%d): deferred(PA)\n",
1117 (unsigned int)handler,signal));
1118 store_signal_data_for_later(data,handler,signal,info,context);
1119 arch_set_pseudo_atomic_interrupted(context);
1120 check_interrupt_context_or_lose(context);
1123 FSHOW_SIGNAL((stderr,
1124 "/maybe_defer_handler(%x,%d): not deferred\n",
1125 (unsigned int)handler,signal));
1130 store_signal_data_for_later (struct interrupt_data *data, void *handler,
1132 siginfo_t *info, os_context_t *context)
1134 if (data->pending_handler)
1135 lose("tried to overwrite pending interrupt handler %x with %x\n",
1136 data->pending_handler, handler);
1138 lose("tried to defer null interrupt handler\n");
1139 data->pending_handler = handler;
1140 data->pending_signal = signal;
1142 memcpy(&(data->pending_info), info, sizeof(siginfo_t));
1144 FSHOW_SIGNAL((stderr, "/store_signal_data_for_later: signal: %d\n",
1148 lose("Null context");
1150 /* the signal mask in the context (from before we were
1151 * interrupted) is copied to be restored when run_deferred_handler
1152 * happens. Then the usually-blocked signals are added to the mask
1153 * in the context so that we are running with blocked signals when
1154 * the handler returns */
1155 sigcopyset(&(data->pending_mask),os_context_sigmask_addr(context));
1156 sigaddset_deferrable(os_context_sigmask_addr(context));
1160 maybe_now_maybe_later(int signal, siginfo_t *info, void *void_context)
1162 SAVE_ERRNO(signal,context,void_context);
1163 struct thread *thread = arch_os_get_current_thread();
1164 struct interrupt_data *data = thread->interrupt_data;
1165 if(!maybe_defer_handler(interrupt_handle_now,data,signal,info,context))
1166 interrupt_handle_now(signal, info, context);
1171 low_level_interrupt_handle_now(int signal, siginfo_t *info,
1172 os_context_t *context)
1174 /* No FP control fixage needed, caller has done that. */
1175 check_blockables_blocked_or_lose(0);
1176 check_interrupts_enabled_or_lose(context);
1177 (*interrupt_low_level_handlers[signal])(signal, info, context);
1178 /* No Darwin context fixage needed, caller does that. */
1182 low_level_maybe_now_maybe_later(int signal, siginfo_t *info, void *void_context)
1184 SAVE_ERRNO(signal,context,void_context);
1185 struct thread *thread = arch_os_get_current_thread();
1186 struct interrupt_data *data = thread->interrupt_data;
1188 if(!maybe_defer_handler(low_level_interrupt_handle_now,data,
1189 signal,info,context))
1190 low_level_interrupt_handle_now(signal, info, context);
1195 #ifdef LISP_FEATURE_SB_THREAD
1197 /* This function must not cons, because that may trigger a GC. */
1199 sig_stop_for_gc_handler(int signal, siginfo_t *info, os_context_t *context)
1201 struct thread *thread=arch_os_get_current_thread();
1202 boolean was_in_lisp;
1204 /* Test for GC_INHIBIT _first_, else we'd trap on every single
1205 * pseudo atomic until gc is finally allowed. */
1206 if (SymbolValue(GC_INHIBIT,thread) != NIL) {
1207 FSHOW_SIGNAL((stderr, "sig_stop_for_gc deferred (*GC-INHIBIT*)\n"));
1208 SetSymbolValue(STOP_FOR_GC_PENDING,T,thread);
1210 } else if (arch_pseudo_atomic_atomic(context)) {
1211 FSHOW_SIGNAL((stderr,"sig_stop_for_gc deferred (PA)\n"));
1212 SetSymbolValue(STOP_FOR_GC_PENDING,T,thread);
1213 arch_set_pseudo_atomic_interrupted(context);
1214 maybe_save_gc_mask_and_block_deferrables
1215 (os_context_sigmask_addr(context));
1219 FSHOW_SIGNAL((stderr, "/sig_stop_for_gc_handler\n"));
1221 /* Not PA and GC not inhibited -- we can stop now. */
1223 was_in_lisp = !foreign_function_call_active_p(arch_os_get_current_thread());
1226 /* need the context stored so it can have registers scavenged */
1227 fake_foreign_function_call(context);
1230 /* Not pending anymore. */
1231 SetSymbolValue(GC_PENDING,NIL,thread);
1232 SetSymbolValue(STOP_FOR_GC_PENDING,NIL,thread);
1234 /* Consider this: in a PA section GC is requested: GC_PENDING,
1235 * pseudo_atomic_interrupted and gc_blocked_deferrables are set,
1236 * deferrables are blocked then pseudo_atomic_atomic is cleared,
1237 * but a SIG_STOP_FOR_GC arrives before trapping to
1238 * interrupt_handle_pending. Here, GC_PENDING is cleared but
1239 * pseudo_atomic_interrupted is not and we go on running with
1240 * pseudo_atomic_interrupted but without a pending interrupt or
1241 * GC. GC_BLOCKED_DEFERRABLES is also left at 1. So let's tidy it
1243 if (thread->interrupt_data->gc_blocked_deferrables) {
1244 FSHOW_SIGNAL((stderr,"cleaning up after gc_blocked_deferrables\n"));
1245 clear_pseudo_atomic_interrupted(thread);
1246 sigcopyset(os_context_sigmask_addr(context),
1247 &thread->interrupt_data->pending_mask);
1248 thread->interrupt_data->gc_blocked_deferrables = 0;
1251 if(thread_state(thread)!=STATE_RUNNING) {
1252 lose("sig_stop_for_gc_handler: wrong thread state: %ld\n",
1253 fixnum_value(thread->state));
1256 set_thread_state(thread,STATE_STOPPED);
1257 FSHOW_SIGNAL((stderr,"suspended\n"));
1259 /* While waiting for gc to finish occupy ourselves with zeroing
1260 * the unused portion of the control stack to reduce conservatism.
1261 * On hypothetic platforms with threads and exact gc it is
1262 * actually a must. */
1263 scrub_control_stack();
1265 wait_for_thread_state_change(thread, STATE_STOPPED);
1266 FSHOW_SIGNAL((stderr,"resumed\n"));
1268 if(thread_state(thread)!=STATE_RUNNING) {
1269 lose("sig_stop_for_gc_handler: wrong thread state on wakeup: %ld\n",
1270 fixnum_value(thread_state(thread)));
1274 undo_fake_foreign_function_call(context);
1281 interrupt_handle_now_handler(int signal, siginfo_t *info, void *void_context)
1283 SAVE_ERRNO(signal,context,void_context);
1284 #ifndef LISP_FEATURE_WIN32
1285 if ((signal == SIGILL) || (signal == SIGBUS)
1286 #ifndef LISP_FEATURE_LINUX
1287 || (signal == SIGEMT)
1290 corruption_warning_and_maybe_lose("Signal %d received", signal);
1292 interrupt_handle_now(signal, info, context);
1296 /* manipulate the signal context and stack such that when the handler
1297 * returns, it will call function instead of whatever it was doing
1301 #if (defined(LISP_FEATURE_X86) || defined(LISP_FEATURE_X86_64))
1302 extern int *context_eflags_addr(os_context_t *context);
1305 extern lispobj call_into_lisp(lispobj fun, lispobj *args, int nargs);
1306 extern void post_signal_tramp(void);
1307 extern void call_into_lisp_tramp(void);
1309 arrange_return_to_lisp_function(os_context_t *context, lispobj function)
1311 #ifndef LISP_FEATURE_WIN32
1312 check_gc_signals_unblocked_or_lose
1313 (os_context_sigmask_addr(context));
1315 #if !(defined(LISP_FEATURE_X86) || defined(LISP_FEATURE_X86_64))
1316 void * fun=native_pointer(function);
1317 void *code = &(((struct simple_fun *) fun)->code);
1320 /* Build a stack frame showing `interrupted' so that the
1321 * user's backtrace makes (as much) sense (as usual) */
1323 /* fp state is saved and restored by call_into_lisp */
1324 /* FIXME: errno is not restored, but since current uses of this
1325 * function only call Lisp code that signals an error, it's not
1326 * much of a problem. In other words, running out of the control
1327 * stack between a syscall and (GET-ERRNO) may clobber errno if
1328 * something fails during signalling or in the handler. But I
1329 * can't see what can go wrong as long as there is no CONTINUE
1330 * like restart on them. */
1331 #ifdef LISP_FEATURE_X86
1332 /* Suppose the existence of some function that saved all
1333 * registers, called call_into_lisp, then restored GP registers and
1334 * returned. It would look something like this:
1342 pushl {address of function to call}
1343 call 0x8058db0 <call_into_lisp>
1350 * What we do here is set up the stack that call_into_lisp would
1351 * expect to see if it had been called by this code, and frob the
1352 * signal context so that signal return goes directly to call_into_lisp,
1353 * and when that function (and the lisp function it invoked) returns,
1354 * it returns to the second half of this imaginary function which
1355 * restores all registers and returns to C
1357 * For this to work, the latter part of the imaginary function
1358 * must obviously exist in reality. That would be post_signal_tramp
1361 u32 *sp=(u32 *)*os_context_register_addr(context,reg_ESP);
1363 #if defined(LISP_FEATURE_DARWIN)
1364 u32 *register_save_area = (u32 *)os_validate(0, 0x40);
1366 FSHOW_SIGNAL((stderr, "/arrange_return_to_lisp_function: preparing to go to function %x, sp: %x\n", function, sp));
1367 FSHOW_SIGNAL((stderr, "/arrange_return_to_lisp_function: context: %x, &context %x\n", context, &context));
1369 /* 1. os_validate (malloc/mmap) register_save_block
1370 * 2. copy register state into register_save_block
1371 * 3. put a pointer to register_save_block in a register in the context
1372 * 4. set the context's EIP to point to a trampoline which:
1373 * a. builds the fake stack frame from the block
1374 * b. frees the block
1375 * c. calls the function
1378 *register_save_area = *os_context_pc_addr(context);
1379 *(register_save_area + 1) = function;
1380 *(register_save_area + 2) = *os_context_register_addr(context,reg_EDI);
1381 *(register_save_area + 3) = *os_context_register_addr(context,reg_ESI);
1382 *(register_save_area + 4) = *os_context_register_addr(context,reg_EDX);
1383 *(register_save_area + 5) = *os_context_register_addr(context,reg_ECX);
1384 *(register_save_area + 6) = *os_context_register_addr(context,reg_EBX);
1385 *(register_save_area + 7) = *os_context_register_addr(context,reg_EAX);
1386 *(register_save_area + 8) = *context_eflags_addr(context);
1388 *os_context_pc_addr(context) =
1389 (os_context_register_t) call_into_lisp_tramp;
1390 *os_context_register_addr(context,reg_ECX) =
1391 (os_context_register_t) register_save_area;
1394 /* return address for call_into_lisp: */
1395 *(sp-15) = (u32)post_signal_tramp;
1396 *(sp-14) = function; /* args for call_into_lisp : function*/
1397 *(sp-13) = 0; /* arg array */
1398 *(sp-12) = 0; /* no. args */
1399 /* this order matches that used in POPAD */
1400 *(sp-11)=*os_context_register_addr(context,reg_EDI);
1401 *(sp-10)=*os_context_register_addr(context,reg_ESI);
1403 *(sp-9)=*os_context_register_addr(context,reg_ESP)-8;
1404 /* POPAD ignores the value of ESP: */
1406 *(sp-7)=*os_context_register_addr(context,reg_EBX);
1408 *(sp-6)=*os_context_register_addr(context,reg_EDX);
1409 *(sp-5)=*os_context_register_addr(context,reg_ECX);
1410 *(sp-4)=*os_context_register_addr(context,reg_EAX);
1411 *(sp-3)=*context_eflags_addr(context);
1412 *(sp-2)=*os_context_register_addr(context,reg_EBP);
1413 *(sp-1)=*os_context_pc_addr(context);
1417 #elif defined(LISP_FEATURE_X86_64)
1418 u64 *sp=(u64 *)*os_context_register_addr(context,reg_RSP);
1420 /* return address for call_into_lisp: */
1421 *(sp-18) = (u64)post_signal_tramp;
1423 *(sp-17)=*os_context_register_addr(context,reg_R15);
1424 *(sp-16)=*os_context_register_addr(context,reg_R14);
1425 *(sp-15)=*os_context_register_addr(context,reg_R13);
1426 *(sp-14)=*os_context_register_addr(context,reg_R12);
1427 *(sp-13)=*os_context_register_addr(context,reg_R11);
1428 *(sp-12)=*os_context_register_addr(context,reg_R10);
1429 *(sp-11)=*os_context_register_addr(context,reg_R9);
1430 *(sp-10)=*os_context_register_addr(context,reg_R8);
1431 *(sp-9)=*os_context_register_addr(context,reg_RDI);
1432 *(sp-8)=*os_context_register_addr(context,reg_RSI);
1433 /* skip RBP and RSP */
1434 *(sp-7)=*os_context_register_addr(context,reg_RBX);
1435 *(sp-6)=*os_context_register_addr(context,reg_RDX);
1436 *(sp-5)=*os_context_register_addr(context,reg_RCX);
1437 *(sp-4)=*os_context_register_addr(context,reg_RAX);
1438 *(sp-3)=*context_eflags_addr(context);
1439 *(sp-2)=*os_context_register_addr(context,reg_RBP);
1440 *(sp-1)=*os_context_pc_addr(context);
1442 *os_context_register_addr(context,reg_RDI) =
1443 (os_context_register_t)function; /* function */
1444 *os_context_register_addr(context,reg_RSI) = 0; /* arg. array */
1445 *os_context_register_addr(context,reg_RDX) = 0; /* no. args */
1447 struct thread *th=arch_os_get_current_thread();
1448 build_fake_control_stack_frames(th,context);
1451 #ifdef LISP_FEATURE_X86
1453 #if !defined(LISP_FEATURE_DARWIN)
1454 *os_context_pc_addr(context) = (os_context_register_t)call_into_lisp;
1455 *os_context_register_addr(context,reg_ECX) = 0;
1456 *os_context_register_addr(context,reg_EBP) = (os_context_register_t)(sp-2);
1458 *os_context_register_addr(context,reg_UESP) =
1459 (os_context_register_t)(sp-15);
1461 *os_context_register_addr(context,reg_ESP) = (os_context_register_t)(sp-15);
1462 #endif /* __NETBSD__ */
1463 #endif /* LISP_FEATURE_DARWIN */
1465 #elif defined(LISP_FEATURE_X86_64)
1466 *os_context_pc_addr(context) = (os_context_register_t)call_into_lisp;
1467 *os_context_register_addr(context,reg_RCX) = 0;
1468 *os_context_register_addr(context,reg_RBP) = (os_context_register_t)(sp-2);
1469 *os_context_register_addr(context,reg_RSP) = (os_context_register_t)(sp-18);
1471 /* this much of the calling convention is common to all
1473 *os_context_pc_addr(context) = (os_context_register_t)(unsigned long)code;
1474 *os_context_register_addr(context,reg_NARGS) = 0;
1475 *os_context_register_addr(context,reg_LIP) =
1476 (os_context_register_t)(unsigned long)code;
1477 *os_context_register_addr(context,reg_CFP) =
1478 (os_context_register_t)(unsigned long)access_control_frame_pointer(th);
1480 #ifdef ARCH_HAS_NPC_REGISTER
1481 *os_context_npc_addr(context) =
1482 4 + *os_context_pc_addr(context);
1484 #ifdef LISP_FEATURE_SPARC
1485 *os_context_register_addr(context,reg_CODE) =
1486 (os_context_register_t)(fun + FUN_POINTER_LOWTAG);
1488 FSHOW((stderr, "/arranged return to Lisp function (0x%lx)\n",
1492 /* KLUDGE: Theoretically the approach we use for undefined alien
1493 * variables should work for functions as well, but on PPC/Darwin
1494 * we get bus error at bogus addresses instead, hence this workaround,
1495 * that has the added benefit of automatically discriminating between
1496 * functions and variables.
1499 undefined_alien_function(void)
1501 funcall0(StaticSymbolFunction(UNDEFINED_ALIEN_FUNCTION_ERROR));
1504 void lower_thread_control_stack_guard_page(struct thread *th)
1506 protect_control_stack_guard_page(0, th);
1507 protect_control_stack_return_guard_page(1, th);
1508 th->control_stack_guard_page_protected = NIL;
1509 fprintf(stderr, "INFO: Control stack guard page unprotected\n");
1512 void reset_thread_control_stack_guard_page(struct thread *th)
1514 memset(CONTROL_STACK_GUARD_PAGE(th), 0, os_vm_page_size);
1515 protect_control_stack_guard_page(1, th);
1516 protect_control_stack_return_guard_page(0, th);
1517 th->control_stack_guard_page_protected = T;
1518 fprintf(stderr, "INFO: Control stack guard page reprotected\n");
1521 /* Called from the REPL, too. */
1522 void reset_control_stack_guard_page(void)
1524 struct thread *th=arch_os_get_current_thread();
1525 if (th->control_stack_guard_page_protected == NIL) {
1526 reset_thread_control_stack_guard_page(th);
1530 void lower_control_stack_guard_page(void)
1532 lower_thread_control_stack_guard_page(arch_os_get_current_thread());
1536 handle_guard_page_triggered(os_context_t *context,os_vm_address_t addr)
1538 struct thread *th=arch_os_get_current_thread();
1540 if(addr >= CONTROL_STACK_HARD_GUARD_PAGE(th) &&
1541 addr < CONTROL_STACK_HARD_GUARD_PAGE(th) + os_vm_page_size) {
1542 lose("Control stack exhausted");
1544 else if(addr >= CONTROL_STACK_GUARD_PAGE(th) &&
1545 addr < CONTROL_STACK_GUARD_PAGE(th) + os_vm_page_size) {
1546 /* We hit the end of the control stack: disable guard page
1547 * protection so the error handler has some headroom, protect the
1548 * previous page so that we can catch returns from the guard page
1549 * and restore it. */
1550 if (th->control_stack_guard_page_protected == NIL)
1551 lose("control_stack_guard_page_protected NIL");
1552 lower_control_stack_guard_page();
1553 #ifdef LISP_FEATURE_C_STACK_IS_CONTROL_STACK
1554 /* For the unfortunate case, when the control stack is
1555 * exhausted in a signal handler. */
1556 unblock_signals_in_context_and_maybe_warn(context);
1558 arrange_return_to_lisp_function
1559 (context, StaticSymbolFunction(CONTROL_STACK_EXHAUSTED_ERROR));
1562 else if(addr >= CONTROL_STACK_RETURN_GUARD_PAGE(th) &&
1563 addr < CONTROL_STACK_RETURN_GUARD_PAGE(th) + os_vm_page_size) {
1564 /* We're returning from the guard page: reprotect it, and
1565 * unprotect this one. This works even if we somehow missed
1566 * the return-guard-page, and hit it on our way to new
1567 * exhaustion instead. */
1568 if (th->control_stack_guard_page_protected != NIL)
1569 lose("control_stack_guard_page_protected not NIL");
1570 reset_control_stack_guard_page();
1573 else if(addr >= BINDING_STACK_HARD_GUARD_PAGE(th) &&
1574 addr < BINDING_STACK_HARD_GUARD_PAGE(th) + os_vm_page_size) {
1575 lose("Binding stack exhausted");
1577 else if(addr >= BINDING_STACK_GUARD_PAGE(th) &&
1578 addr < BINDING_STACK_GUARD_PAGE(th) + os_vm_page_size) {
1579 protect_binding_stack_guard_page(0, NULL);
1580 protect_binding_stack_return_guard_page(1, NULL);
1581 fprintf(stderr, "INFO: Binding stack guard page unprotected\n");
1583 /* For the unfortunate case, when the binding stack is
1584 * exhausted in a signal handler. */
1585 unblock_signals_in_context_and_maybe_warn(context);
1586 arrange_return_to_lisp_function
1587 (context, StaticSymbolFunction(BINDING_STACK_EXHAUSTED_ERROR));
1590 else if(addr >= BINDING_STACK_RETURN_GUARD_PAGE(th) &&
1591 addr < BINDING_STACK_RETURN_GUARD_PAGE(th) + os_vm_page_size) {
1592 protect_binding_stack_guard_page(1, NULL);
1593 protect_binding_stack_return_guard_page(0, NULL);
1594 fprintf(stderr, "INFO: Binding stack guard page reprotected\n");
1597 else if(addr >= ALIEN_STACK_HARD_GUARD_PAGE(th) &&
1598 addr < ALIEN_STACK_HARD_GUARD_PAGE(th) + os_vm_page_size) {
1599 lose("Alien stack exhausted");
1601 else if(addr >= ALIEN_STACK_GUARD_PAGE(th) &&
1602 addr < ALIEN_STACK_GUARD_PAGE(th) + os_vm_page_size) {
1603 protect_alien_stack_guard_page(0, NULL);
1604 protect_alien_stack_return_guard_page(1, NULL);
1605 fprintf(stderr, "INFO: Alien stack guard page unprotected\n");
1607 /* For the unfortunate case, when the alien stack is
1608 * exhausted in a signal handler. */
1609 unblock_signals_in_context_and_maybe_warn(context);
1610 arrange_return_to_lisp_function
1611 (context, StaticSymbolFunction(ALIEN_STACK_EXHAUSTED_ERROR));
1614 else if(addr >= ALIEN_STACK_RETURN_GUARD_PAGE(th) &&
1615 addr < ALIEN_STACK_RETURN_GUARD_PAGE(th) + os_vm_page_size) {
1616 protect_alien_stack_guard_page(1, NULL);
1617 protect_alien_stack_return_guard_page(0, NULL);
1618 fprintf(stderr, "INFO: Alien stack guard page reprotected\n");
1621 else if (addr >= undefined_alien_address &&
1622 addr < undefined_alien_address + os_vm_page_size) {
1623 arrange_return_to_lisp_function
1624 (context, StaticSymbolFunction(UNDEFINED_ALIEN_VARIABLE_ERROR));
1631 * noise to install handlers
1634 #ifndef LISP_FEATURE_WIN32
1635 /* In Linux 2.4 synchronous signals (sigtrap & co) can be delivered if
1636 * they are blocked, in Linux 2.6 the default handler is invoked
1637 * instead that usually coredumps. One might hastily think that adding
1638 * SA_NODEFER helps, but until ~2.6.13 if SA_NODEFER is specified then
1639 * the whole sa_mask is ignored and instead of not adding the signal
1640 * in question to the mask. That means if it's not blockable the
1641 * signal must be unblocked at the beginning of signal handlers.
1643 * It turns out that NetBSD's SA_NODEFER doesn't DTRT in a different
1644 * way: if SA_NODEFER is set and the signal is in sa_mask, the signal
1645 * will be unblocked in the sigmask during the signal handler. -- RMK
1648 static volatile int sigaction_nodefer_works = -1;
1650 #define SA_NODEFER_TEST_BLOCK_SIGNAL SIGABRT
1651 #define SA_NODEFER_TEST_KILL_SIGNAL SIGUSR1
1654 sigaction_nodefer_test_handler(int signal, siginfo_t *info, void *void_context)
1658 get_current_sigmask(¤t);
1659 /* There should be exactly two blocked signals: the two we added
1660 * to sa_mask when setting up the handler. NetBSD doesn't block
1661 * the signal we're handling when SA_NODEFER is set; Linux before
1662 * 2.6.13 or so also doesn't block the other signal when
1663 * SA_NODEFER is set. */
1664 for(i = 1; i < NSIG; i++)
1665 if (sigismember(¤t, i) !=
1666 (((i == SA_NODEFER_TEST_BLOCK_SIGNAL) || (i == signal)) ? 1 : 0)) {
1667 FSHOW_SIGNAL((stderr, "SA_NODEFER doesn't work, signal %d\n", i));
1668 sigaction_nodefer_works = 0;
1670 if (sigaction_nodefer_works == -1)
1671 sigaction_nodefer_works = 1;
1675 see_if_sigaction_nodefer_works(void)
1677 struct sigaction sa, old_sa;
1679 sa.sa_flags = SA_SIGINFO | SA_NODEFER;
1680 sa.sa_sigaction = sigaction_nodefer_test_handler;
1681 sigemptyset(&sa.sa_mask);
1682 sigaddset(&sa.sa_mask, SA_NODEFER_TEST_BLOCK_SIGNAL);
1683 sigaddset(&sa.sa_mask, SA_NODEFER_TEST_KILL_SIGNAL);
1684 sigaction(SA_NODEFER_TEST_KILL_SIGNAL, &sa, &old_sa);
1685 /* Make sure no signals are blocked. */
1688 sigemptyset(&empty);
1689 thread_sigmask(SIG_SETMASK, &empty, 0);
1691 kill(getpid(), SA_NODEFER_TEST_KILL_SIGNAL);
1692 while (sigaction_nodefer_works == -1);
1693 sigaction(SA_NODEFER_TEST_KILL_SIGNAL, &old_sa, NULL);
1696 #undef SA_NODEFER_TEST_BLOCK_SIGNAL
1697 #undef SA_NODEFER_TEST_KILL_SIGNAL
1700 unblock_me_trampoline(int signal, siginfo_t *info, void *void_context)
1702 SAVE_ERRNO(signal,context,void_context);
1705 sigemptyset(&unblock);
1706 sigaddset(&unblock, signal);
1707 thread_sigmask(SIG_UNBLOCK, &unblock, 0);
1708 interrupt_handle_now(signal, info, context);
1713 low_level_unblock_me_trampoline(int signal, siginfo_t *info, void *void_context)
1715 SAVE_ERRNO(signal,context,void_context);
1718 sigemptyset(&unblock);
1719 sigaddset(&unblock, signal);
1720 thread_sigmask(SIG_UNBLOCK, &unblock, 0);
1721 (*interrupt_low_level_handlers[signal])(signal, info, context);
1726 low_level_handle_now_handler(int signal, siginfo_t *info, void *void_context)
1728 SAVE_ERRNO(signal,context,void_context);
1729 (*interrupt_low_level_handlers[signal])(signal, info, context);
1734 undoably_install_low_level_interrupt_handler (int signal,
1735 interrupt_handler_t handler)
1737 struct sigaction sa;
1739 if (0 > signal || signal >= NSIG) {
1740 lose("bad signal number %d\n", signal);
1743 if (ARE_SAME_HANDLER(handler, SIG_DFL))
1744 sa.sa_sigaction = (void (*)(int, siginfo_t*, void*))handler;
1745 else if (sigismember(&deferrable_sigset,signal))
1746 sa.sa_sigaction = low_level_maybe_now_maybe_later;
1747 else if (!sigaction_nodefer_works &&
1748 !sigismember(&blockable_sigset, signal))
1749 sa.sa_sigaction = low_level_unblock_me_trampoline;
1751 sa.sa_sigaction = low_level_handle_now_handler;
1753 sigcopyset(&sa.sa_mask, &blockable_sigset);
1754 sa.sa_flags = SA_SIGINFO | SA_RESTART
1755 | (sigaction_nodefer_works ? SA_NODEFER : 0);
1756 #ifdef LISP_FEATURE_C_STACK_IS_CONTROL_STACK
1757 if(signal==SIG_MEMORY_FAULT)
1758 sa.sa_flags |= SA_ONSTACK;
1761 sigaction(signal, &sa, NULL);
1762 interrupt_low_level_handlers[signal] =
1763 (ARE_SAME_HANDLER(handler, SIG_DFL) ? 0 : handler);
1767 /* This is called from Lisp. */
1769 install_handler(int signal, void handler(int, siginfo_t*, os_context_t*))
1771 #ifndef LISP_FEATURE_WIN32
1772 struct sigaction sa;
1774 union interrupt_handler oldhandler;
1776 FSHOW((stderr, "/entering POSIX install_handler(%d, ..)\n", signal));
1778 block_blockable_signals(0, &old);
1780 FSHOW((stderr, "/interrupt_low_level_handlers[signal]=%x\n",
1781 (unsigned int)interrupt_low_level_handlers[signal]));
1782 if (interrupt_low_level_handlers[signal]==0) {
1783 if (ARE_SAME_HANDLER(handler, SIG_DFL) ||
1784 ARE_SAME_HANDLER(handler, SIG_IGN))
1785 sa.sa_sigaction = (void (*)(int, siginfo_t*, void*))handler;
1786 else if (sigismember(&deferrable_sigset, signal))
1787 sa.sa_sigaction = maybe_now_maybe_later;
1788 else if (!sigaction_nodefer_works &&
1789 !sigismember(&blockable_sigset, signal))
1790 sa.sa_sigaction = unblock_me_trampoline;
1792 sa.sa_sigaction = interrupt_handle_now_handler;
1794 sigcopyset(&sa.sa_mask, &blockable_sigset);
1795 sa.sa_flags = SA_SIGINFO | SA_RESTART |
1796 (sigaction_nodefer_works ? SA_NODEFER : 0);
1797 sigaction(signal, &sa, NULL);
1800 oldhandler = interrupt_handlers[signal];
1801 interrupt_handlers[signal].c = handler;
1803 thread_sigmask(SIG_SETMASK, &old, 0);
1805 FSHOW((stderr, "/leaving POSIX install_handler(%d, ..)\n", signal));
1807 return (unsigned long)oldhandler.lisp;
1809 /* Probably-wrong Win32 hack */
1814 /* This must not go through lisp as it's allowed anytime, even when on
1817 sigabrt_handler(int signal, siginfo_t *info, os_context_t *context)
1819 lose("SIGABRT received.\n");
1823 interrupt_init(void)
1825 #ifndef LISP_FEATURE_WIN32
1827 SHOW("entering interrupt_init()");
1828 see_if_sigaction_nodefer_works();
1829 sigemptyset(&deferrable_sigset);
1830 sigemptyset(&blockable_sigset);
1831 sigemptyset(&gc_sigset);
1832 sigaddset_deferrable(&deferrable_sigset);
1833 sigaddset_blockable(&blockable_sigset);
1834 sigaddset_gc(&gc_sigset);
1836 /* Set up high level handler information. */
1837 for (i = 0; i < NSIG; i++) {
1838 interrupt_handlers[i].c =
1839 /* (The cast here blasts away the distinction between
1840 * SA_SIGACTION-style three-argument handlers and
1841 * signal(..)-style one-argument handlers, which is OK
1842 * because it works to call the 1-argument form where the
1843 * 3-argument form is expected.) */
1844 (void (*)(int, siginfo_t*, os_context_t*))SIG_DFL;
1846 undoably_install_low_level_interrupt_handler(SIGABRT, sigabrt_handler);
1847 SHOW("returning from interrupt_init()");
1851 #ifndef LISP_FEATURE_WIN32
1853 siginfo_code(siginfo_t *info)
1855 return info->si_code;
1857 os_vm_address_t current_memory_fault_address;
1860 lisp_memory_fault_error(os_context_t *context, os_vm_address_t addr)
1862 /* FIXME: This is lossy: if we get another memory fault (eg. from
1863 * another thread) before lisp has read this, we lose the information.
1864 * However, since this is mostly informative, we'll live with that for
1865 * now -- some address is better then no address in this case.
1867 current_memory_fault_address = addr;
1868 /* To allow debugging memory faults in signal handlers and such. */
1869 corruption_warning_and_maybe_lose("Memory fault at %x (pc=%p, sp=%p)",
1871 *os_context_pc_addr(context),
1872 #ifdef ARCH_HAS_STACK_POINTER
1873 *os_context_sp_addr(context)
1878 unblock_signals_in_context_and_maybe_warn(context);
1879 #ifdef LISP_FEATURE_C_STACK_IS_CONTROL_STACK
1880 arrange_return_to_lisp_function(context,
1881 StaticSymbolFunction(MEMORY_FAULT_ERROR));
1883 funcall0(StaticSymbolFunction(MEMORY_FAULT_ERROR));
1889 unhandled_trap_error(os_context_t *context)
1891 lispobj context_sap;
1892 fake_foreign_function_call(context);
1893 unblock_gc_signals(0, 0);
1894 context_sap = alloc_sap(context);
1895 #ifndef LISP_FEATURE_WIN32
1896 thread_sigmask(SIG_SETMASK, os_context_sigmask_addr(context), 0);
1898 funcall1(StaticSymbolFunction(UNHANDLED_TRAP_ERROR), context_sap);
1899 lose("UNHANDLED-TRAP-ERROR fell through");
1902 /* Common logic for trapping instructions. How we actually handle each
1903 * case is highly architecture dependent, but the overall shape is
1906 handle_trap(os_context_t *context, int trap)
1909 case trap_PendingInterrupt:
1910 FSHOW((stderr, "/<trap pending interrupt>\n"));
1911 arch_skip_instruction(context);
1912 interrupt_handle_pending(context);
1916 FSHOW((stderr, "/<trap error/cerror %d>\n", trap));
1917 interrupt_internal_error(context, trap==trap_Cerror);
1919 case trap_Breakpoint:
1920 arch_handle_breakpoint(context);
1922 case trap_FunEndBreakpoint:
1923 arch_handle_fun_end_breakpoint(context);
1925 #ifdef trap_AfterBreakpoint
1926 case trap_AfterBreakpoint:
1927 arch_handle_after_breakpoint(context);
1930 #ifdef trap_SingleStepAround
1931 case trap_SingleStepAround:
1932 case trap_SingleStepBefore:
1933 arch_handle_single_step_trap(context, trap);
1937 fake_foreign_function_call(context);
1938 lose("%%PRIMITIVE HALT called; the party is over.\n");
1940 unhandled_trap_error(context);