We are observing an issue with a property-based test triggering an apparent infinite loop on arm64 (5.0.0~alpha0 and 5.1.0+trunk)
while amd64 completes the test as expected.
Consider the property-based test in https://github.com/jmid/multicoretests/tree/bytes located in file src/bytes/lin_tests.ml
Starting from a Bytes of size 42, this repeats 1000 times a test of the following form:
|
Bytes.get t 4082802875569992363
Bytes.get t -2858709553352776116
Bytes.get t 1363109893947027853
Bytes.get t 2947451885998816648
|
.------------------------------------.
| |
Bytes.get t 489138274756058094 Bytes.get t 4420947024389089565
| Bytes.get t -1978859559514483936
| |
spawns two Domains and runs a number calls in parallel.Bytes.get - hence shouldn't trigger a read/write or write/write race.Invalid_argument.I just did a little experiment:
caml_scan_stack (the longest run I interrupted after at least 30min). This is also the behaviour observed on a MacBook Pro M1 - similarly with an ARM64 CPU.We are observing the same behaviour with OCaml version 5.0.0~alpha0 and latest 5.1.0+trunk.
perf top reveals that the test is spending most its time in caml_scan_stack:
92.86% lin_tests.exe [.] caml_scan_stack
0.23% [kernel] [k] _raw_spin_unlock_irq
0.22% [kernel] [k] clear_page
0.21% perf_5.10 [.] 0x00000000000f846c
0.20% libc-2.31.so [.] __aarch64_swp4_rel
0.19% [kernel] [k] kallsyms_expand_symbol.constprop.7
0.16% perf_5.10 [.] 0x00000000001024d4
0.16% [kernel] [k] _raw_spin_unlock_irqrestore
0.14% [kernel] [k] vsnprintf
0.13% [kernel] [k] module_get_kallsym
0.12% [kernel] [k] el0_svc_common.constprop.3
0.12% [kernel] [k] format_decode
0.11% ld-2.31.so [.] do_lookup_x
0.10% libc-2.31.so [.] _int_malloc
0.10% [kernel] [k] number
0.10% [kernel] [k] string_nocheck
0.09% libc-2.31.so [.] __GI___memset_generic
0.09% libc-2.31.so [.] strchr
0.08% [kernel] [k] get_page_from_freelist
0.07% libc-2.31.so [.] __libc_calloc
0.07% [kernel] [k] el0_da
0.07% [kernel] [k] __arch_copy_to_user
0.06% perf_5.10 [.] 0x00000000001024ec
0.05% libc-2.31.so [.] __mprotect
0.05% libelf-0.183.so [.] gelf_getsym
0.05% [kernel] [k] memcpy
0.05% libc-2.31.so [.] strcmp
0.05% libc-2.31.so [.] __memcpy_generic
0.05% libc-2.31.so [.] sysmalloc
0.04% ld-2.31.so [.] _dl_lookup_symbol_x
0.04% [kernel] [k] preempt_count_add
0.04% libc-2.31.so [.] __aarch64_cas4_acq
0.04% libc-2.31.so [.] strrchr
0.04% libslang.so.2.3.2 [.] SLsmg_write_chars
0.03% ld-2.31.so [.] _dl_relocate_object
0.03% libc-2.31.so [.] __strlen_generic
0.03% perf_5.10 [.] 0x0000000000102a0c
0.03% [kernel] [k] ___bpf_prog_run
0.03% [kernel] [k] preempt_count_sub
0.03% perf_5.10 [.] 0x00000000001024e4
0.03% [kernel] [k] generic_file_write_iter
0.03% perf_5.10 [.] 0x00000000000ed124
0.03% [kernel] [k] prep_new_page
0.03% [kernel] [k] handle_mm_fault
0.03% libc-2.31.so [.] __aarch64_cas8_rel
Here's the perf report for a folded run where I interrupted the run after a while:
$ perf report --stdio -n -g folded
# To display the perf.data header info, please use --header/--header-only options.
#
#
# Total Lost Samples: 0
#
# Samples: 230K of event 'cycles:u'
# Event count (approx.): 94799659549
#
# Children Self Samples Command Shared Object Symbol
# ........ ........ ............ ............. .................. ..........................................................................................
#
99.31% 0.02% 116 lin_tests.exe lin_tests.exe [.] stw_handler
98.87% stw_handler;caml_stw_empty_minor_heap;caml_stw_empty_minor_heap_no_major_slice.constprop.0;caml_empty_minor_heap_promote;caml_scan_stack
99.01% 0.00% 3 lin_tests.exe lin_tests.exe [.] caml_stw_empty_minor_heap
98.87% caml_stw_empty_minor_heap;caml_stw_empty_minor_heap_no_major_slice.constprop.0;caml_empty_minor_heap_promote;caml_scan_stack
99.00% 0.01% 36 lin_tests.exe lin_tests.exe [.] caml_stw_empty_minor_heap_no_major_slice.constprop.0
98.87% caml_stw_empty_minor_heap_no_major_slice.constprop.0;caml_empty_minor_heap_promote;caml_scan_stack
98.97% 0.01% 24 lin_tests.exe lin_tests.exe [.] caml_empty_minor_heap_promote
98.87% caml_empty_minor_heap_promote;caml_scan_stack
98.91% 0.02% 186 lin_tests.exe lin_tests.exe [.] caml_c_call
98.87% caml_c_call;caml_array_bound_error;caml_raise;caml_process_pending_actions_with_root_exn;caml_do_pending_actions_exn;caml_handle_gc_interrupt;stw_handler;caml>
98.87% 98.87% 212408 lin_tests.exe lin_tests.exe [.] caml_scan_stack
98.87% caml_c_call;caml_array_bound_error;caml_raise;caml_process_pending_actions_with_root_exn;caml_do_pending_actions_exn;caml_handle_gc_interrupt;stw_handler;caml>
98.87% 0.00% 0 lin_tests.exe lin_tests.exe [.] caml_do_pending_actions_exn
98.87% caml_do_pending_actions_exn;caml_handle_gc_interrupt;stw_handler;caml_stw_empty_minor_heap;caml_stw_empty_minor_heap_no_major_slice.constprop.0;caml_empty_min>
98.87% 0.00% 0 lin_tests.exe lin_tests.exe [.] caml_handle_gc_interrupt
98.87% caml_handle_gc_interrupt;stw_handler;caml_stw_empty_minor_heap;caml_stw_empty_minor_heap_no_major_slice.constprop.0;caml_empty_minor_heap_promote;caml_scan_st>
98.87% 0.00% 1 lin_tests.exe lin_tests.exe [.] caml_process_pending_actions_with_root_exn
98.87% caml_process_pending_actions_with_root_exn;caml_do_pending_actions_exn;caml_handle_gc_interrupt;stw_handler;caml_stw_empty_minor_heap;caml_stw_empty_minor_hea>
98.87% 0.00% 4 lin_tests.exe lin_tests.exe [.] caml_array_bound_error
98.87% caml_array_bound_error;caml_raise;caml_process_pending_actions_with_root_exn;caml_do_pending_actions_exn;caml_handle_gc_interrupt;stw_handler;caml_stw_empty_m>
98.87% 0.00% 4 lin_tests.exe lin_tests.exe [.] caml_raise
98.87% caml_raise;caml_process_pending_actions_with_root_exn;caml_do_pending_actions_exn;caml_handle_gc_interrupt;stw_handler;caml_stw_empty_minor_heap;caml_stw_empt>
0.80% 0.00% 1870 lin_tests.exe libc-2.31.so [.] thread_start
0.80% 0.00% 4931 lin_tests.exe libpthread-2.31.so [.] start_thread
0.43% 0.00% 352 lin_tests.exe lin_tests.exe [.] backup_thread_func
0.36% 0.00% 534 lin_tests.exe lin_tests.exe [.] domain_thread_func
0.34% 0.01% 46 lin_tests.exe lin_tests.exe [.] major_collection_slice
0.31% 0.02% 64 lin_tests.exe lin_tests.exe [.] caml_empty_minor_heaps_once
0.26% 0.09% 430 lin_tests.exe lin_tests.exe [.] caml_try_run_on_all_domains_with_spin_work
0.25% 0.25% 895 lin_tests.exe lin_tests.exe [.] pool_sweep
0.25% 0.00% 3 lin_tests.exe lin_tests.exe [.] caml_sweep
0.11% 0.11% 339 lin_tests.exe lin_tests.exe [.] caml_gc_log
0.08% 0.03% 223 lin_tests.exe lin_tests.exe [.] camlLin__fun_2177
0.08% 0.00% 0 lin_tests.exe [unknown] [.] 0x0100000000000005
0.08% 0.01% 43 lin_tests.exe lin_tests.exe [.] oldify_one
0.06% 0.00% 0 dune dune [.] caml_poll_gc_work
To recreate:
git clone -b bytes https://github.com/jmid/multicoretests.git
opam update
opam install qcheck
dune exec -- src/bytes/lin_tests.exe -v
This issue was first reported by @naomiiiiiiiii
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