-
-
Notifications
You must be signed in to change notification settings - Fork 5.5k
/
signals-unix.c
1249 lines (1159 loc) · 42.5 KB
/
signals-unix.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
// This file is a part of Julia. License is MIT: https://julialang.org/license
// Note that this file is `#include`d by "signal-handling.c"
#include <signal.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/mman.h>
#include <pthread.h>
#include <time.h>
#include <errno.h>
#include "julia.h"
#include "julia_internal.h"
#if defined(_OS_DARWIN_) && !defined(MAP_ANONYMOUS)
#define MAP_ANONYMOUS MAP_ANON
#endif
#ifdef __APPLE__
#include <AvailabilityMacros.h>
#ifdef MAC_OS_X_VERSION_10_9
#include <sys/_types/_ucontext64.h>
#else
#define __need_ucontext64_t
#include <machine/_structs.h>
#endif
#endif
// Figure out the best signals/timers to use for this platform
#if defined(__APPLE__) // Darwin's mach ports allow signal-free thread management
#define HAVE_MACH
#define HAVE_KEVENT
#elif defined(__OpenBSD__)
#define HAVE_KEVENT
#else // generic Linux or FreeBSD
#define HAVE_TIMER
#endif
#ifdef HAVE_KEVENT
#include <sys/event.h>
#endif
// 8M signal stack, same as default stack size (though we barely use this)
static const size_t sig_stack_size = 8 * 1024 * 1024;
#include "julia_assert.h"
// helper function for returning the unw_context_t inside a ucontext_t
// (also used by stackwalk.c)
bt_context_t *jl_to_bt_context(void *sigctx) JL_NOTSAFEPOINT
{
#ifdef __APPLE__
return (bt_context_t*)&((ucontext64_t*)sigctx)->uc_mcontext64->__ss;
#elif defined(_CPU_ARM_)
// libunwind does not use `ucontext_t` on ARM.
// `unw_context_t` is a struct of 16 `unsigned long` which should
// have the same layout as the `arm_r0` to `arm_pc` fields in `sigcontext`
ucontext_t *ctx = (ucontext_t*)sigctx;
return (bt_context_t*)&ctx->uc_mcontext.arm_r0;
#else
return (bt_context_t*)sigctx;
#endif
}
static int thread0_exit_count = 0;
static void jl_exit_thread0(int signo, jl_bt_element_t *bt_data, size_t bt_size);
int jl_simulate_longjmp(jl_jmp_buf mctx, bt_context_t *c) JL_NOTSAFEPOINT;
static void jl_longjmp_in_ctx(int sig, void *_ctx, jl_jmp_buf jmpbuf);
#if !defined(_OS_DARWIN_)
static inline uintptr_t jl_get_rsp_from_ctx(const void *_ctx)
{
#if defined(_OS_LINUX_) && defined(_CPU_X86_64_)
const ucontext_t *ctx = (const ucontext_t*)_ctx;
return ctx->uc_mcontext.gregs[REG_RSP];
#elif defined(_OS_LINUX_) && defined(_CPU_X86_)
const ucontext_t *ctx = (const ucontext_t*)_ctx;
return ctx->uc_mcontext.gregs[REG_ESP];
#elif defined(_OS_LINUX_) && defined(_CPU_AARCH64_)
const ucontext_t *ctx = (const ucontext_t*)_ctx;
return ctx->uc_mcontext.sp;
#elif defined(_OS_LINUX_) && defined(_CPU_ARM_)
const ucontext_t *ctx = (const ucontext_t*)_ctx;
return ctx->uc_mcontext.arm_sp;
#elif defined(_OS_LINUX_) && (defined(_CPU_RISCV64_))
const ucontext_t *ctx = (const ucontext_t*)_ctx;
return ctx->uc_mcontext.__gregs[REG_SP];
#elif defined(_OS_FREEBSD_) && defined(_CPU_X86_64_)
const ucontext_t *ctx = (const ucontext_t*)_ctx;
return ctx->uc_mcontext.mc_rsp;
#elif defined(_OS_FREEBSD_) && defined(_CPU_AARCH64_)
const ucontext_t *ctx = (const ucontext_t*)_ctx;
return ctx->uc_mcontext.mc_gpregs.gp_sp;
#elif defined(_OS_OPENBSD_) && defined(_CPU_X86_64_)
const struct sigcontext *ctx = (const struct sigcontext *)_ctx;
return ctx->sc_rsp;
#else
// TODO Add support for PowerPC(64)?
return 0;
#endif
}
static int is_addr_on_sigstack(jl_ptls_t ptls, void *ptr) JL_NOTSAFEPOINT
{
// One guard page for signal_stack.
return ptls->signal_stack == NULL ||
((char*)ptr >= (char*)ptls->signal_stack - jl_page_size &&
(char*)ptr <= (char*)ptls->signal_stack + (ptls->signal_stack_size ? ptls->signal_stack_size : sig_stack_size));
}
// Modify signal context `_ctx` so that `fptr` will execute when the signal returns
// The function `fptr` itself must not return.
JL_NO_ASAN static void jl_call_in_ctx(jl_ptls_t ptls, void (*fptr)(void), int sig, void *_ctx)
{
// Modifying the ucontext should work but there is concern that
// sigreturn oriented programming mitigation can work against us
// by rejecting ucontext that is modified.
// The current (staged) implementation in the Linux Kernel only
// checks that the syscall is made in the signal handler and that
// the ucontext address is valid. Hopefully the value of the ucontext
// will not be part of the validation...
uintptr_t rsp = jl_get_rsp_from_ctx(_ctx);
rsp = (rsp - 256) & ~(uintptr_t)15; // redzone and re-alignment
#if defined(_OS_LINUX_) && defined(_CPU_X86_64_)
ucontext_t *ctx = (ucontext_t*)_ctx;
rsp -= sizeof(void*);
*(uintptr_t*)rsp = 0;
ctx->uc_mcontext.gregs[REG_RSP] = rsp;
ctx->uc_mcontext.gregs[REG_RIP] = (uintptr_t)fptr;
#elif defined(_OS_FREEBSD_) && defined(_CPU_X86_64_)
ucontext_t *ctx = (ucontext_t*)_ctx;
rsp -= sizeof(void*);
*(uintptr_t*)rsp = 0;
ctx->uc_mcontext.mc_rsp = rsp;
ctx->uc_mcontext.mc_rip = (uintptr_t)fptr;
#elif defined(_OS_LINUX_) && defined(_CPU_X86_)
ucontext_t *ctx = (ucontext_t*)_ctx;
rsp -= sizeof(void*);
*(uintptr_t*)rsp = 0;
ctx->uc_mcontext.gregs[REG_ESP] = rsp;
ctx->uc_mcontext.gregs[REG_EIP] = (uintptr_t)fptr;
#elif defined(_OS_FREEBSD_) && defined(_CPU_X86_)
ucontext_t *ctx = (ucontext_t*)_ctx;
rsp -= sizeof(void*);
*(uintptr_t*)rsp = 0;
ctx->uc_mcontext.mc_esp = rsp;
ctx->uc_mcontext.mc_eip = (uintptr_t)fptr;
#elif defined(_OS_OPENBSD_) && defined(_CPU_X86_64_)
struct sigcontext *ctx = (struct sigcontext *)_ctx;
rsp -= sizeof(void*);
*(uintptr_t*)rsp = 0;
ctx->sc_rsp = rsp;
ctx->sc_rip = fptr;
#elif defined(_OS_LINUX_) && defined(_CPU_AARCH64_)
ucontext_t *ctx = (ucontext_t*)_ctx;
ctx->uc_mcontext.sp = rsp;
ctx->uc_mcontext.regs[29] = 0; // Clear link register (x29)
ctx->uc_mcontext.pc = (uintptr_t)fptr;
#elif defined(_OS_FREEBSD_) && defined(_CPU_AARCH64_)
ucontext_t *ctx = (ucontext_t*)_ctx;
ctx->uc_mcontext.mc_gpregs.gp_sp = rsp;
ctx->uc_mcontext.mc_gpregs.gp_x[29] = 0; // Clear link register (x29)
ctx->uc_mcontext.mc_gpregs.gp_elr = (uintptr_t)fptr;
#elif defined(_OS_LINUX_) && defined(_CPU_ARM_)
ucontext_t *ctx = (ucontext_t*)_ctx;
uintptr_t target = (uintptr_t)fptr;
// Apparently some glibc's sigreturn target is running in thumb state.
// Mimic a `bx` instruction by setting the T(5) bit of CPSR
// depending on the target address.
uintptr_t cpsr = ctx->uc_mcontext.arm_cpsr;
// Thumb mode function pointer should have the lowest bit set
if (target & 1) {
target = target & ~((uintptr_t)1);
cpsr = cpsr | (1 << 5);
}
else {
cpsr = cpsr & ~(1 << 5);
}
ctx->uc_mcontext.arm_cpsr = cpsr;
ctx->uc_mcontext.arm_sp = rsp;
ctx->uc_mcontext.arm_lr = 0; // Clear link register
ctx->uc_mcontext.arm_pc = target;
#elif defined(_OS_LINUX_) && (defined(_CPU_RISCV64_))
ucontext_t *ctx = (ucontext_t*)_ctx;
ctx->uc_mcontext.__gregs[REG_SP] = rsp;
ctx->uc_mcontext.__gregs[REG_RA] = 0; // Clear return address address (ra)
ctx->uc_mcontext.__gregs[REG_PC] = (uintptr_t)fptr;
#else
#pragma message("julia: throw-in-context not supported on this platform")
// TODO Add support for PowerPC(64)?
sigset_t sset;
sigemptyset(&sset);
sigaddset(&sset, sig);
pthread_sigmask(SIG_UNBLOCK, &sset, NULL);
fptr();
#endif
}
#endif
static void jl_throw_in_ctx(jl_task_t *ct, jl_value_t *e, int sig, void *sigctx)
{
jl_ptls_t ptls = ct->ptls;
assert(!jl_get_safe_restore());
ptls->bt_size =
rec_backtrace_ctx(ptls->bt_data, JL_MAX_BT_SIZE, jl_to_bt_context(sigctx),
ct->gcstack);
ptls->sig_exception = e;
ptls->io_wait = 0;
jl_handler_t *eh = ct->eh;
if (eh != NULL) {
asan_unpoison_task_stack(ct, &eh->eh_ctx);
jl_longjmp_in_ctx(sig, sigctx, eh->eh_ctx);
}
else {
jl_no_exc_handler(e, ct);
}
}
static pthread_t signals_thread;
static int is_addr_on_stack(jl_task_t *ct, void *addr) JL_NOTSAFEPOINT
{
if (ct->ctx.copy_stack) {
jl_ptls_t ptls = ct->ptls;
return ((char*)addr > (char*)ptls->stackbase - ptls->stacksize &&
(char*)addr < (char*)ptls->stackbase);
}
return ((char*)addr > (char*)ct->ctx.stkbuf &&
(char*)addr < (char*)ct->ctx.stkbuf + ct->ctx.bufsz);
}
static void sigdie_handler(int sig, siginfo_t *info, void *context)
{
signal(sig, SIG_DFL);
uv_tty_reset_mode();
if (sig == SIGILL)
jl_show_sigill(context);
jl_task_t *ct = jl_get_current_task();
jl_critical_error(sig, info->si_code, jl_to_bt_context(context), ct);
if (ct)
jl_atomic_store_relaxed(&ct->ptls->safepoint, (size_t*)NULL + 1);
if (info->si_code == 0 ||
info->si_code == SI_USER ||
#ifdef SI_KERNEL
info->si_code == SI_KERNEL ||
#endif
info->si_code == SI_QUEUE ||
#ifdef SI_MESGQ
info->si_code == SI_MESGQ ||
#endif
#ifdef SI_ASYNCIO
info->si_code == SI_ASYNCIO ||
#endif
#ifdef SI_SIGIO
info->si_code == SI_SIGIO ||
#endif
#ifdef SI_TKILL
info->si_code == SI_TKILL ||
#endif
info->si_code == SI_TIMER)
raise(sig);
else if (sig != SIGSEGV &&
sig != SIGBUS &&
sig != SIGILL &&
sig != SIGFPE &&
sig != SIGTRAP)
raise(sig);
// fall-through return to re-execute faulting statement (but without the
// error handler and the pgcstack having been destroyed)
}
#if defined(_CPU_X86_64_) || defined(_CPU_X86_)
enum x86_trap_flags {
USER_MODE = 0x4,
WRITE_FAULT = 0x2,
PAGE_PRESENT = 0x1 // whether this page is currently mapped into memory
};
int exc_reg_is_write_fault(uintptr_t err) {
return err & WRITE_FAULT;
}
#elif defined(_CPU_AARCH64_)
enum aarch64_esr_layout {
EC_MASK = ((uint32_t)0b111111) << 26,
EC_DATA_ABORT = ((uint32_t)0b100100) << 26,
DFSC_MASK = ((uint32_t)0b111111) << 0,
ISR_DA_WnR = ((uint32_t)1) << 6
};
int exc_reg_is_write_fault(uintptr_t esr) {
// n.b. we check that DFSC is either a permission fault (page in memory but not writable) or a translation fault (page not in memory)
// but because of info->si_code == SEGV_ACCERR, we know the kernel could have brought the page into memory.
// Access faults happen when trying to write to code or secure memory, which is a more severe violation, so we ignore those.
// AArch64 appears to leaves it up to a given implementer whether atomic update errors are reported as read or write faults.
return (esr & EC_MASK) == EC_DATA_ABORT &&
(((esr & DFSC_MASK) >= 0b000100 && // Translation flag fault, level 0.
(esr & DFSC_MASK) <= 0b000111) || // Translation fault, level 3.
((esr & DFSC_MASK) >= 0b001100 && // Permission flag fault, level 0.
(esr & DFSC_MASK) <= 0b001111)) && // Permission fault, level 3.
(esr & ISR_DA_WnR); // Attempted write
}
#endif
#if defined(HAVE_MACH)
#include "signals-mach.c"
#else
#include <poll.h>
#include <sys/eventfd.h>
int jl_lock_stackwalk(void)
{
jl_lock_profile();
return 0;
}
void jl_unlock_stackwalk(int lockret)
{
(void)lockret;
jl_unlock_profile();
}
#if defined(_OS_LINUX_) && (defined(_CPU_X86_64_) || defined(_CPU_X86_))
int is_write_fault(void *context) {
ucontext_t *ctx = (ucontext_t*)context;
return exc_reg_is_write_fault(ctx->uc_mcontext.gregs[REG_ERR]);
}
#elif defined(_OS_LINUX_) && defined(_CPU_AARCH64_)
struct linux_aarch64_ctx_header {
uint32_t magic;
uint32_t size;
};
const uint32_t linux_esr_magic = 0x45535201;
int is_write_fault(void *context) {
ucontext_t *ctx = (ucontext_t*)context;
struct linux_aarch64_ctx_header *extra =
(struct linux_aarch64_ctx_header *)ctx->uc_mcontext.__reserved;
while (extra->magic != 0) {
if (extra->magic == linux_esr_magic) {
return exc_reg_is_write_fault(*(uint64_t*)&extra[1]);
}
extra = (struct linux_aarch64_ctx_header *)
(((uint8_t*)extra) + extra->size);
}
return 0;
}
#elif defined(_OS_FREEBSD_) && (defined(_CPU_X86_64_) || defined(_CPU_X86_))
int is_write_fault(void *context) {
ucontext_t *ctx = (ucontext_t*)context;
return exc_reg_is_write_fault(ctx->uc_mcontext.mc_err);
}
#elif defined(_OS_FREEBSD_) && defined(_CPU_AARCH64_)
// FreeBSD seems not to expose a means of accessing ESR via `ucontext_t` on AArch64.
// TODO: Is there an alternative approach that can be taken? ESR may become accessible
// in a future release though.
int is_write_fault(void *context) {
return 0;
}
#elif defined(_OS_OPENBSD_) && defined(_CPU_X86_64_)
int is_write_fault(void *context) {
struct sigcontext *ctx = (struct sigcontext *)context;
return exc_reg_is_write_fault(ctx->sc_err);
}
#else
#pragma message("Implement this query for consistent PROT_NONE handling")
int is_write_fault(void *context) {
return 0;
}
#endif
static int jl_is_on_sigstack(jl_ptls_t ptls, void *ptr, void *context) JL_NOTSAFEPOINT
{
return (ptls->signal_stack != NULL &&
is_addr_on_sigstack(ptls, ptr) &&
is_addr_on_sigstack(ptls, (void*)jl_get_rsp_from_ctx(context)));
}
JL_NO_ASAN static void segv_handler(int sig, siginfo_t *info, void *context)
{
assert(sig == SIGSEGV || sig == SIGBUS);
jl_jmp_buf *saferestore = jl_get_safe_restore();
if (saferestore) { // restarting jl_ or profile
jl_longjmp_in_ctx(sig, context, *saferestore);
return;
}
jl_task_t *ct = jl_get_current_task();
if (ct == NULL || ct->ptls == NULL || jl_atomic_load_relaxed(&ct->ptls->gc_state) == JL_GC_STATE_WAITING) {
sigdie_handler(sig, info, context);
return;
}
if (sig == SIGSEGV && info->si_code == SEGV_ACCERR && jl_addr_is_safepoint((uintptr_t)info->si_addr) && !is_write_fault(context)) {
jl_set_gc_and_wait();
// Do not raise sigint on worker thread
if (jl_atomic_load_relaxed(&ct->tid) != 0)
return;
// n.b. if the user might have seen that we were in a state where it
// was safe to run GC concurrently, we might briefly enter a state
// where our execution is not consistent with the gc_state of this
// thread. That will quickly be rectified when we rerun the faulting
// instruction and end up right back here, or we start to run the
// exception handler and immediately hit the safepoint there.
if (ct->ptls->defer_signal) {
jl_safepoint_defer_sigint();
}
else if (jl_safepoint_consume_sigint()) {
jl_clear_force_sigint();
jl_throw_in_ctx(ct, jl_interrupt_exception, sig, context);
}
return;
}
if (ct->eh == NULL)
sigdie_handler(sig, info, context);
if ((sig != SIGBUS || info->si_code == BUS_ADRERR) && is_addr_on_stack(ct, info->si_addr)) { // stack overflow and not a BUS_ADRALN (alignment error)
stack_overflow_warning();
jl_throw_in_ctx(ct, jl_stackovf_exception, sig, context);
}
else if (jl_is_on_sigstack(ct->ptls, info->si_addr, context)) {
// This mainly happens when one of the finalizers during final cleanup
// on the signal stack has a deep/infinite recursion.
// There isn't anything more we can do
// (we are already corrupting that stack running this function)
// so just call `_exit` to terminate immediately.
jl_safe_printf("ERROR: Signal stack overflow, exit\n");
jl_raise(sig);
}
else if (sig == SIGSEGV && info->si_code == SEGV_ACCERR && is_write_fault(context)) { // writing to read-only memory (e.g., mmap)
jl_throw_in_ctx(ct, jl_readonlymemory_exception, sig, context);
}
else {
sigdie_handler(sig, info, context);
}
}
pthread_mutex_t in_signal_lock; // shared with jl_delete_thread
static bt_context_t *signal_context; // protected by in_signal_lock
static int exit_signal_cond = -1;
static int signal_caught_cond = -1;
static int signals_inflight = 0;
int jl_thread_suspend_and_get_state(int tid, int timeout, bt_context_t *ctx)
{
int err;
pthread_mutex_lock(&in_signal_lock);
jl_ptls_t ptls2 = jl_atomic_load_relaxed(&jl_all_tls_states)[tid];
jl_task_t *ct2 = ptls2 ? jl_atomic_load_relaxed(&ptls2->current_task) : NULL;
if (ct2 == NULL) {
// this thread is not alive or already dead
pthread_mutex_unlock(&in_signal_lock);
return 0;
}
while (signals_inflight) {
// something is wrong, or there is already a usr2 in flight elsewhere
// try to wait for it to finish or wait for timeout
struct pollfd event = {signal_caught_cond, POLLIN, 0};
do {
err = poll(&event, 1, timeout * 1000);
} while (err == -1 && errno == EINTR);
if (err == -1 || (event.revents & POLLIN) == 0) {
// not ready after timeout: cancel this request
pthread_mutex_unlock(&in_signal_lock);
return 0;
}
// consume it before continuing
eventfd_t got;
do {
err = read(signal_caught_cond, &got, sizeof(eventfd_t));
} while (err == -1 && errno == EINTR);
if (err != sizeof(eventfd_t)) abort();
assert(signals_inflight >= got);
signals_inflight -= got;
}
signals_inflight++;
sig_atomic_t request = jl_atomic_exchange(&ptls2->signal_request, 1);
assert(request == 0 || request == -1);
request = 1;
err = pthread_kill(ptls2->system_id, SIGUSR2);
if (err == 0) {
// wait for thread to acknowledge or timeout
struct pollfd event = {signal_caught_cond, POLLIN, 0};
do {
err = poll(&event, 1, timeout * 1000);
} while (err == -1 && errno == EINTR);
if (err != 1 || (event.revents & POLLIN) == 0)
err = -1;
}
if (err == -1) {
// not ready after timeout: try to cancel this request
if (jl_atomic_cmpswap(&ptls2->signal_request, &request, 0)) {
signals_inflight--;
pthread_mutex_unlock(&in_signal_lock);
return 0;
}
}
eventfd_t got;
do {
err = read(signal_caught_cond, &got, sizeof(eventfd_t));
} while (err == -1 && errno == EINTR);
if (err != sizeof(eventfd_t)) abort();
assert(signals_inflight >= got);
signals_inflight -= got;
signals_inflight++;
// Now the other thread is waiting on exit_signal_cond (verify that here by
// checking it is 0, and add an acquire barrier for good measure)
request = jl_atomic_load_acquire(&ptls2->signal_request);
assert(request == 0 || request == -1); (void) request;
jl_atomic_store_release(&ptls2->signal_request, 4); // prepare to resume normally, but later code may change this
*ctx = *signal_context;
return 1;
}
void jl_thread_resume(int tid)
{
int err;
eventfd_t got = 1;
err = write(exit_signal_cond, &got, sizeof(eventfd_t));
if (err != sizeof(eventfd_t)) abort();
pthread_mutex_unlock(&in_signal_lock);
}
// Throw jl_interrupt_exception if the master thread is in a signal async region
// or if SIGINT happens too often.
static void jl_try_deliver_sigint(void)
{
jl_ptls_t ptls2 = jl_atomic_load_relaxed(&jl_all_tls_states)[0];
jl_safepoint_enable_sigint();
jl_wake_libuv();
pthread_mutex_lock(&in_signal_lock);
signals_inflight++;
jl_atomic_store_release(&ptls2->signal_request, 2);
// This also makes sure `sleep` is aborted.
pthread_kill(ptls2->system_id, SIGUSR2);
pthread_mutex_unlock(&in_signal_lock);
}
// Write only by signal handling thread, read only by main thread
// no sync necessary.
static int thread0_exit_signo = 0;
static void JL_NORETURN jl_exit_thread0_cb(void)
{
CFI_NORETURN
jl_critical_error(thread0_exit_signo, 0, NULL, jl_current_task);
jl_atexit_hook(128);
jl_raise(thread0_exit_signo);
}
static void jl_exit_thread0(int signo, jl_bt_element_t *bt_data, size_t bt_size)
{
jl_ptls_t ptls2 = jl_atomic_load_relaxed(&jl_all_tls_states)[0];
bt_context_t signal_context;
// This also makes sure `sleep` is aborted.
if (jl_thread_suspend_and_get_state(0, 30, &signal_context)) {
thread0_exit_signo = signo;
ptls2->bt_size = bt_size; // <= JL_MAX_BT_SIZE
memcpy(ptls2->bt_data, bt_data, ptls2->bt_size * sizeof(bt_data[0]));
jl_atomic_store_release(&ptls2->signal_request, 3);
jl_thread_resume(0); // resume with message 3 (call jl_exit_thread0_cb)
}
else {
// thread 0 is gone? just do the exit ourself
jl_raise(signo);
}
}
// request:
// -1: processing
// 0: nothing [not from here]
// 1: get state & wait for request
// 2: throw sigint if `!defer_signal && io_wait` or if force throw threshold
// is reached
// 3: raise `thread0_exit_signo` and try to exit
// 4: no-op
void usr2_handler(int sig, siginfo_t *info, void *ctx)
{
jl_task_t *ct = jl_get_current_task();
if (ct == NULL)
return;
jl_ptls_t ptls = ct->ptls;
if (ptls == NULL)
return;
int errno_save = errno;
sig_atomic_t request = jl_atomic_load(&ptls->signal_request);
if (request == 0)
return;
if (!jl_atomic_cmpswap(&ptls->signal_request, &request, -1))
return;
if (request == 1) {
signal_context = jl_to_bt_context(ctx);
// acknowledge that we saw the signal_request and set signal_context
int err;
eventfd_t got = 1;
err = write(signal_caught_cond, &got, sizeof(eventfd_t));
if (err != sizeof(eventfd_t)) abort();
sig_atomic_t processing = -1;
jl_atomic_cmpswap(&ptls->signal_request, &processing, 0);
// wait for exit signal
do {
err = read(exit_signal_cond, &got, sizeof(eventfd_t));
} while (err == -1 && errno == EINTR);
if (err != sizeof(eventfd_t)) abort();
assert(got == 1);
request = jl_atomic_exchange(&ptls->signal_request, -1);
signal_context = NULL;
assert(request == 2 || request == 3 || request == 4);
}
int err;
eventfd_t got = 1;
err = write(signal_caught_cond, &got, sizeof(eventfd_t));
if (err != sizeof(eventfd_t)) abort();
sig_atomic_t processing = -1;
jl_atomic_cmpswap(&ptls->signal_request, &processing, 0);
if (request == 2) {
int force = jl_check_force_sigint();
if (force || (!ptls->defer_signal && ptls->io_wait)) {
jl_safepoint_consume_sigint();
if (force)
jl_safe_printf("WARNING: Force throwing a SIGINT\n");
// Force a throw
jl_clear_force_sigint();
jl_jmp_buf *saferestore = jl_get_safe_restore();
if (saferestore) // restarting jl_ or profile
jl_longjmp_in_ctx(sig, ctx, *saferestore);
else
jl_throw_in_ctx(ct, jl_interrupt_exception, sig, ctx);
}
}
else if (request == 3) {
jl_call_in_ctx(ct->ptls, jl_exit_thread0_cb, sig, ctx);
}
errno = errno_save;
}
// Because SIGUSR1 is dual-purpose, and the timer can have trailing signals after being deleted,
// a 2-second grace period is imposed to ignore any trailing timer-created signals so they don't get
// confused for user triggers
uint64_t last_timer_delete_time = 0;
int timer_graceperiod_elapsed(void)
{
return jl_hrtime() > (last_timer_delete_time + 2e9);
}
#if defined(HAVE_TIMER)
// Linux-style
#include <time.h>
#include <string.h> // for memset
static timer_t timerprof;
static struct itimerspec itsprof;
JL_DLLEXPORT int jl_profile_start_timer(uint8_t all_tasks)
{
struct sigevent sigprof;
// Establish the signal event
memset(&sigprof, 0, sizeof(struct sigevent));
sigprof.sigev_notify = SIGEV_SIGNAL;
sigprof.sigev_signo = SIGUSR1;
sigprof.sigev_value.sival_ptr = &timerprof;
// Because SIGUSR1 is multipurpose, set `profile_running` before so that we know that the first SIGUSR1 came from the timer
profile_running = 1;
profile_all_tasks = all_tasks;
if (timer_create(CLOCK_REALTIME, &sigprof, &timerprof) == -1) {
profile_running = 0;
profile_all_tasks = 0;
return -2;
}
// Start the timer
itsprof.it_interval.tv_sec = 0;
itsprof.it_interval.tv_nsec = 0;
itsprof.it_value.tv_sec = nsecprof / GIGA;
itsprof.it_value.tv_nsec = nsecprof % GIGA;
if (timer_settime(timerprof, 0, &itsprof, NULL) == -1) {
profile_running = 0;
profile_all_tasks = 0;
return -3;
}
return 0;
}
JL_DLLEXPORT void jl_profile_stop_timer(void)
{
uv_mutex_lock(&bt_data_prof_lock);
if (profile_running) {
timer_delete(timerprof);
last_timer_delete_time = jl_hrtime();
profile_running = 0;
}
uv_mutex_unlock(&bt_data_prof_lock);
}
#elif defined(__OpenBSD__)
JL_DLLEXPORT int jl_profile_start_timer(void)
{
return -1;
}
JL_DLLEXPORT void jl_profile_stop_timer(void)
{
}
#else
#error no profile tools available
#endif
#endif // HAVE_MACH
static void allocate_segv_handler(void)
{
struct sigaction act;
memset(&act, 0, sizeof(struct sigaction));
sigemptyset(&act.sa_mask);
act.sa_sigaction = segv_handler;
act.sa_flags = SA_ONSTACK | SA_SIGINFO;
if (sigaction(SIGSEGV, &act, NULL) < 0) {
jl_errorf("fatal error: sigaction: %s", strerror(errno));
}
// On AArch64, stack overflow triggers a SIGBUS
if (sigaction(SIGBUS, &act, NULL) < 0) {
jl_errorf("fatal error: sigaction: %s", strerror(errno));
}
}
void jl_install_thread_signal_handler(jl_ptls_t ptls)
{
#ifdef HAVE_MACH
attach_exception_port(pthread_mach_thread_np(ptls->system_id), 0);
#endif
stack_t ss;
if (sigaltstack(NULL, &ss) < 0)
jl_errorf("fatal error: sigaltstack: %s", strerror(errno));
if ((ss.ss_flags & SS_DISABLE) != SS_DISABLE)
return; // someone else appears to have already set this up, so just use that
size_t ssize = sig_stack_size;
void *signal_stack = jl_malloc_stack(&ssize, NULL);
ss.ss_flags = 0;
ss.ss_size = ssize;
assert(ssize != 0);
#ifndef _OS_OPENBSD_
/* fallback to malloc(), but it isn't possible on OpenBSD */
if (signal_stack == NULL) {
signal_stack = malloc(ssize);
ssize = 0;
if (signal_stack == NULL)
jl_safe_printf("\nwarning: julia signal alt stack could not be allocated (StackOverflowError will be fatal on this thread).\n");
else
jl_safe_printf("\nwarning: julia signal stack allocated without guard page (launch foreign threads earlier to avoid this warning).\n");
}
#endif
if (signal_stack != NULL) {
ss.ss_sp = signal_stack;
if (sigaltstack(&ss, NULL) < 0)
jl_errorf("fatal error: sigaltstack: %s", strerror(errno));
ptls->signal_stack = signal_stack;
ptls->signal_stack_size = ssize;
}
}
const static int sigwait_sigs[] = {
SIGINT, SIGTERM, SIGQUIT,
#ifdef SIGINFO
SIGINFO,
#else
SIGUSR1,
#endif
#if defined(HAVE_TIMER)
SIGUSR1,
#endif
0
};
static void jl_sigsetset(sigset_t *sset)
{
sigemptyset(sset);
for (const int *sig = sigwait_sigs; *sig; sig++)
sigaddset(sset, *sig);
}
#ifdef HAVE_KEVENT
static void kqueue_signal(int *sigqueue, struct kevent *ev, int sig)
{
if (*sigqueue == -1)
return;
EV_SET(ev, sig, EVFILT_SIGNAL, EV_ADD, 0, 0, 0);
if (kevent(*sigqueue, ev, 1, NULL, 0, NULL)) {
perror("signal kevent");
close(*sigqueue);
*sigqueue = -1;
}
else {
// kqueue gets signals before SIG_IGN, but does not remove them from pending (unlike sigwait)
signal(sig, SIG_IGN);
}
}
#endif
void trigger_profile_peek(void)
{
jl_safe_printf("\n======================================================================================\n");
jl_safe_printf("Information request received. A stacktrace will print followed by a %.1f second profile\n", profile_peek_duration);
jl_safe_printf("======================================================================================\n");
if (profile_bt_size_max == 0){
// If the buffer hasn't been initialized, initialize with default size
// Keep these values synchronized with Profile.default_init()
if (jl_profile_init(10000000, 1000000) == -1) {
jl_safe_printf("ERROR: could not initialize the profile buffer");
return;
}
}
profile_bt_size_cur = 0; // clear profile buffer
if (jl_profile_start_timer(0) < 0)
jl_safe_printf("ERROR: Could not start profile timer\n");
else
profile_autostop_time = jl_hrtime() + (profile_peek_duration * 1e9);
}
// assumes holding `jl_lock_stackwalk`
void jl_profile_thread_unix(int tid, bt_context_t *signal_context)
{
if (jl_profile_is_buffer_full()) {
// Buffer full: Delete the timer
jl_profile_stop_timer();
return;
}
// notify thread to stop
if (!jl_thread_suspend_and_get_state(tid, 1, signal_context))
return;
// unwinding can fail, so keep track of the current state
// and restore from the SEGV handler if anything happens.
jl_jmp_buf *old_buf = jl_get_safe_restore();
jl_jmp_buf buf;
jl_set_safe_restore(&buf);
if (jl_setjmp(buf, 0)) {
jl_safe_printf("WARNING: profiler attempt to access an invalid memory location\n");
} else {
// Get backtrace data
profile_bt_size_cur += rec_backtrace_ctx((jl_bt_element_t*)profile_bt_data_prof + profile_bt_size_cur,
profile_bt_size_max - profile_bt_size_cur - 1, signal_context, NULL);
}
jl_set_safe_restore(old_buf);
jl_ptls_t ptls2 = jl_atomic_load_relaxed(&jl_all_tls_states)[tid];
// store threadid but add 1 as 0 is preserved to indicate end of block
profile_bt_data_prof[profile_bt_size_cur++].uintptr = ptls2->tid + 1;
// store task id (never null)
profile_bt_data_prof[profile_bt_size_cur++].jlvalue = (jl_value_t*)jl_atomic_load_relaxed(&ptls2->current_task);
// store cpu cycle clock
profile_bt_data_prof[profile_bt_size_cur++].uintptr = cycleclock();
// store whether thread is sleeping (don't ever encode a state as `0` since is preserved to indicate end of block)
int state = jl_atomic_load_relaxed(&ptls2->sleep_check_state) == 0 ? PROFILE_STATE_THREAD_NOT_SLEEPING : PROFILE_STATE_THREAD_SLEEPING;
profile_bt_data_prof[profile_bt_size_cur++].uintptr = state;
// Mark the end of this block with two 0's
profile_bt_data_prof[profile_bt_size_cur++].uintptr = 0;
profile_bt_data_prof[profile_bt_size_cur++].uintptr = 0;
// notify thread to resume
jl_thread_resume(tid);
}
static void *signal_listener(void *arg)
{
static jl_bt_element_t bt_data[JL_MAX_BT_SIZE + 1];
static size_t bt_size = 0;
sigset_t sset;
int sig, critical, profile;
jl_sigsetset(&sset);
#if defined(_POSIX_C_SOURCE) && _POSIX_C_SOURCE >= 199309L
siginfo_t info;
#endif
#ifdef HAVE_KEVENT
struct kevent ev;
int sigqueue = kqueue();
if (sigqueue == -1) {
perror("signal kqueue");
}
else {
for (const int *sig = sigwait_sigs; *sig; sig++)
kqueue_signal(&sigqueue, &ev, *sig);
if (sigqueue == -1) {
// re-enable sigwait for these
for (const int *sig = sigwait_sigs; *sig; sig++)
signal(*sig, SIG_DFL);
}
}
#endif
while (1) {
sig = 0;
errno = 0;
#ifdef HAVE_KEVENT
if (sigqueue != -1) {
int nevents = kevent(sigqueue, NULL, 0, &ev, 1, NULL);
if (nevents == -1) {
if (errno == EINTR)
continue;
perror("signal kevent");
}
if (nevents != 1) {
close(sigqueue);
sigqueue = -1;
for (const int *sig = sigwait_sigs; *sig; sig++)
signal(*sig, SIG_DFL);
continue;
}
sig = ev.ident;
}
else
#endif
#if defined(_POSIX_C_SOURCE) && _POSIX_C_SOURCE >= 199309L
sig = sigwaitinfo(&sset, &info);
#else
if (sigwait(&sset, &sig))
sig = -1;
#endif
if (sig == -1) {
if (errno == EINTR)
continue;
sig = SIGABRT; // this branch can't occur, unless we had stack memory corruption of sset
}
profile = 0;
#ifndef HAVE_MACH
#if defined(HAVE_TIMER)
profile = (sig == SIGUSR1);
#if defined(_POSIX_C_SOURCE) && _POSIX_C_SOURCE >= 199309L
if (profile && !(info.si_code == SI_TIMER &&
info.si_value.sival_ptr == &timerprof))
profile = 0;
#endif
#endif
#endif
if (sig == SIGINT) {
if (jl_ignore_sigint()) {
continue;
}
else if (exit_on_sigint) {
critical = 1;
}
else {
jl_try_deliver_sigint();
continue;
}
}
else {
critical = 0;
}
critical |= (sig == SIGTERM);
critical |= (sig == SIGABRT);
critical |= (sig == SIGQUIT);
#ifdef SIGINFO
critical |= (sig == SIGINFO);
#else
critical |= (sig == SIGUSR1 && !profile);
#endif
int doexit = critical;
#ifdef SIGINFO
if (sig == SIGINFO) {
if (profile_running != 1)
trigger_profile_peek();
doexit = 0;
}
#else
if (sig == SIGUSR1) {
if (profile_running != 1 && timer_graceperiod_elapsed())
trigger_profile_peek();
doexit = 0;
}
#endif
if (doexit) {
// The exit can get stuck if it happens at an unfortunate spot in thread 0
// (unavoidable due to its async nature).
// Try much harder to exit next time, if we get multiple exit requests.
// 1. unblock the signal, so this thread can be killed by it
// 2. reset the tty next, because we might die before we get another chance to do that
// 3. attempt a graceful cleanup of julia, followed by an abrupt end to the C runtime (except for fflush)
// 4. kill this thread with `raise`, to preserve the signo / exit code / and coredump configuration
// Similar to jl_raise, but a slightly different order of operations
sigset_t sset;
sigemptyset(&sset);
sigaddset(&sset, sig);
pthread_sigmask(SIG_UNBLOCK, &sset, NULL);
#ifdef HAVE_KEVENT
signal(sig, SIG_DFL);
#endif
uv_tty_reset_mode();
thread0_exit_count++;