On multicoretests we have observed crashes and deadlocks on macOS ARM64 when running tests performing random allocations and random operations from the Gc-module. We've observed these on 5.2.0, 5.3.0, and on current trunk from today. Interestingly, the misbehaviour doesn't trigger when omitting Gc.compact.
The shape of the tests is the usual: a sequential prefix of random calls, followed by two Domain.spawns each performing random calls too.
In between each iteration, after Domain.join'ning the two domains, we reset the allocation roots and invoke Gc.full_major in an attempt to reset the heap to a somewhat sane state. lldb backtraces reveal that the crashes may happen during this GC - with only the main domain running (see example below).
To recreate:
$ opam switch create . --empty
$ opam install . --inplace-build # build a local opam switch from, e.g., trunk
$ opam install qcheck-core
$ git clone -b gc-test-enable-compact https://github.com/ocaml-multicore/multicoretests.git
$ cd multicoretests
$ dune build
$ dune exec src/gc/stm_tests_par_stress.exe -- -v
Here's an example backtrace from lldb with the main thread #1 encountering an invalid address during gc_full_major while the backup thread #2 is just waiting:
(lldb) bt all
warning: could not execute support code to read Objective-C class data in the process. This may reduce the quality of type information available.
* thread #1, stop reason = ESR_EC_DABORT_EL0 (fault address: 0x100d62c68)
* frame #0: 0x0000000100add134 stm_tests_par_stress.exe`do_some_marking [inlined] Hd_val(val=4309003376) at mlvalues.h:190:10 [opt]
frame #1: 0x0000000100add134 stm_tests_par_stress.exe`do_some_marking(stk=0x00006000038702a0, budget=11590) at major_gc.c:1222:21 [opt]
frame #2: 0x0000000100adba90 stm_tests_par_stress.exe`mark(budget=16384) at major_gc.c:1335:14 [opt]
frame #3: 0x0000000100adb0e4 stm_tests_par_stress.exe`major_collection_slice(howmuch=<unavailable>, participant_count=1, barrier_participants=0x000000012f008e00, mode=Slice_uninterruptible, force_compaction=0) at major_gc.c:1817:21 [opt]
frame #4: 0x0000000100adb970 stm_tests_par_stress.exe`stw_finish_major_cycle(domain=0x000000012ee047c0, arg=0x000000016f412f40, participating_count=1, participating=0x000000012f008e00) at major_gc.c:2050:5 [opt]
frame #5: 0x0000000100ac8c9c stm_tests_par_stress.exe`caml_try_run_on_all_domains_with_spin_work(sync=<unavailable>, handler=(stm_tests_par_stress.exe`stw_finish_major_cycle at major_gc.c:2030), data=0x000000016f412f40, leader_setup=<unavailable>, enter_spin_callback=<unavailable>, enter_spin_data=<unavailable>) at domain.c:1716:3 [opt]
frame #6: 0x0000000100ac7aec stm_tests_par_stress.exe`caml_try_run_on_all_domains(handler=<unavailable>, data=<unavailable>, leader_setup=<unavailable>) at domain.c:1738:7 [opt] [artificial]
frame #7: 0x0000000100ada944 stm_tests_par_stress.exe`caml_finish_major_cycle(force_compaction=0) at major_gc.c:2065:5 [opt]
frame #8: 0x0000000100ad13d4 stm_tests_par_stress.exe`gc_full_major_res at gc_ctrl.c:269:5 [opt]
frame #9: 0x0000000100ad1390 stm_tests_par_stress.exe`caml_gc_full_major(v=<unavailable>) at gc_ctrl.c:282:34 [opt]
frame #10: 0x0000000100af069c stm_tests_par_stress.exe`caml_c_call + 52
frame #11: 0x00000001009f8298 stm_tests_par_stress.exe`camlDune__exe__Stm_tests_spec$cleanup_1670 + 176
frame #12: 0x00000001009fbe94 stm_tests_par_stress.exe`camlSTM_domain$run_par_660 + 372
frame #13: 0x00000001009fc300 stm_tests_par_stress.exe`camlSTM_domain$stress_prop_par_793 + 32
frame #14: 0x0000000100a01ac8 stm_tests_par_stress.exe`camlUtil$fun_2474 + 120
frame #15: 0x0000000100a191d8 stm_tests_par_stress.exe`camlQCheck2$loop_4000 + 64
frame #16: 0x0000000100a190f8 stm_tests_par_stress.exe`camlQCheck2$run_law_3995 + 104
frame #17: 0x0000000100a19eec stm_tests_par_stress.exe`camlQCheck2$check_state_input_4063 + 204
frame #18: 0x0000000100a1a344 stm_tests_par_stress.exe`camlQCheck2$check_cell_inner_9740 + 316
frame #19: 0x0000000100a09ebc stm_tests_par_stress.exe`camlQCheck_base_runner$aux_map_1444 + 772
frame #20: 0x0000000100a57ba8 stm_tests_par_stress.exe`camlStdlib__List$map_334 + 72
frame #21: 0x0000000100a097cc stm_tests_par_stress.exe`camlQCheck_base_runner$run_tests_inner_2444 + 596
frame #22: 0x0000000100a0a550 stm_tests_par_stress.exe`camlQCheck_base_runner$run_tests_main_inner_3052 + 176
frame #23: 0x00000001009f6138 stm_tests_par_stress.exe`camlDune__exe__Stm_tests_par_stress$entry + 304
frame #24: 0x00000001009f0354 stm_tests_par_stress.exe`caml_program + 1596
frame #25: 0x0000000100af07dc stm_tests_par_stress.exe`caml_start_program + 132
frame #26: 0x0000000100af0058 stm_tests_par_stress.exe`caml_startup_common(argv=0x000000016f4132f0, pooling=<unavailable>) at startup_nat.c:127:9 [opt]
frame #27: 0x0000000100af00bc stm_tests_par_stress.exe`caml_main [inlined] caml_startup_exn(argv=<unavailable>) at startup_nat.c:134:10 [opt]
frame #28: 0x0000000100af00b4 stm_tests_par_stress.exe`caml_main [inlined] caml_startup(argv=<unavailable>) at startup_nat.c:139:15 [opt]
frame #29: 0x0000000100af00b4 stm_tests_par_stress.exe`caml_main(argv=<unavailable>) at startup_nat.c:146:3 [opt]
frame #30: 0x0000000100ada3f4 stm_tests_par_stress.exe`main(argc=<unavailable>, argv=<unavailable>) at main.c:37:3 [opt]
frame #31: 0x00000001875aa0e0 dyld`start + 2360
thread #2
frame #0: 0x00000001878f59ec libsystem_kernel.dylib`__psynch_cvwait + 8
frame #1: 0x000000018793355c libsystem_pthread.dylib`_pthread_cond_wait + 1228
frame #2: 0x0000000100ae4abc stm_tests_par_stress.exe`caml_plat_wait(cond=<unavailable>, mut=<unavailable>) at platform.c:146:21 [opt]
frame #3: 0x0000000100ac9ca8 stm_tests_par_stress.exe`backup_thread_func(v=0x0000000130048000) at domain.c:1080:11 [opt]
RetroSearch is an open source project built by @garambo | Open a GitHub Issue
Search and Browse the WWW like it's 1997 | Search results from DuckDuckGo
HTML:
3.2
| Encoding:
UTF-8
| Version:
0.7.4