[Q] MALLOC_HOOKS default mode description

[hiroyuki-sato]

Hello, Developers.

Could you tell me about the MALLOC_HOOKS description?

The MALLOC_HOOKS description says below.

This is unsafe and off by default,

It seems that this value is on(yes) by default. Does this description correct?

  {"MALLOC_RELOC", "yes",

ucx/src/ucs/config/ucm_opts.c

Lines 57 to 62 in 9050b1a

	{"MALLOC_RELOC", "yes",
	"Enable installing malloc symbols in the relocation table.\n"
	"This is unsafe and off by default, because sometimes glibc\n"
	"calls malloc/free without going through the relocation table,\n"
	"which would use the original implementation and not ours.",
	ucs_offsetof(ucm_global_config_t, enable_malloc_reloc), UCS_CONFIG_TYPE_BOOL},

The default value is zero in this file.

ucx/src/ucm/util/sys.c

Line 38 in 9050b1a

.enable_malloc_reloc = 0,

Best regards

[hiroyuki-sato]

It seems that on by default.

Source: 571381c

ucx_info -c
UCX_MEM_MALLOC_RELOC=y

./ucx_info -f
#
# Enable installing malloc symbols in the relocation table.
# This is unsafe and off by default, because sometimes glibc
# calls malloc/free without going through the relocation table,
# which would use the original implementation and not ours.
#
# syntax:    <y|n>
#
UCX_MEM_MALLOC_RELOC=y

[nmorey]

This should probably be fixed. I'm getting a deadlock with it enabled:
#0 0x00007f9eef53fc07 in sched_yield () from /lib64/libc.so.6
#1 0x00007f9eed3c320e in ?? () from /usr/lib64/libucm.so.0
#2 0x00007f9eed3c483f in ucm_dlmalloc () from /usr/lib64/libucm.so.0
#3 0x00007f9eed3bc24d in ?? () from /usr/lib64/libucm.so.0
#4 0x00007f9ef056a7f8 in _dl_exception_create_format () from /lib64/ld-linux-x86-64.so.2
#5 0x00007f9ef055e91a in _dl_lookup_symbol_x () from /lib64/ld-linux-x86-64.so.2
#6 0x00007f9eef594561 in do_sym () from /lib64/libc.so.6
#7 0x00007f9eede672c4 in dlsym_doit () from /lib64/libdl.so.2
#8 0x00007f9eef594b4d in _dl_catch_exception () from /lib64/libc.so.6
#9 0x00007f9eef594bcf in _dl_catch_error () from /lib64/libc.so.6
#10 0x00007f9eede678a5 in _dlerror_run () from /lib64/libdl.so.2
#11 0x00007f9eede6732f in dlsym () from /lib64/libdl.so.2
#12 0x00007f9eed3be3e9 in ?? () from /usr/lib64/libucm.so.0
#13 0x00007f9eed3bece7 in ucm_orig_dlsym_sbrk () from /usr/lib64/libucm.so.0
#14 0x00007f9eed3b90f3 in ?? () from /usr/lib64/libucm.so.0
#15 0x00007f9eed3b9169 in ucm_event_dispatch () from /usr/lib64/libucm.so.0
#16 0x00007f9eed3ba546 in ucm_sbrk () from /usr/lib64/libucm.so.0
#17 0x00007f9eed3c4b80 in ucm_dlmalloc () from /usr/lib64/libucm.so.0
#18 0x00007f9eed3bc24d in ?? () from /usr/lib64/libucm.so.0
#19 0x00007f9eed3bb1ae in ?? () from /usr/lib64/libucm.so.0
#20 0x00007f9eed3bcb1c in ucm_malloc_install () from /usr/lib64/libucm.so.0
#21 0x00007f9eed3bab9a in ucm_set_event_handler () from /usr/lib64/libucm.so.0
#22 0x00007f9eee4bca43 in ?? () from /usr/lib64/libucs.so.0
#23 0x00007f9eee4be8fd in ucs_rcache_create () from /usr/lib64/libucs.so.0
#24 0x00007f9eed152593 in uct_ib_md_open_common () from /usr/lib64/ucx/libuct_ib.so.0
#25 0x00007f9eed1585c1 in ?? () from /usr/lib64/ucx/libuct_ib.so.0
#26 0x00007f9eed151e34 in uct_ib_md_open () from /usr/lib64/ucx/libuct_ib.so.0
#27 0x00007f9eed7eca12 in uct_md_open () from /usr/lib64/libuct.so.0
#28 0x00007f9eee6fb84c in ?? () from /usr/lib64/libucp.so.0
#29 0x00007f9eee6fd101 in ucp_init_version () from /usr/lib64/libucp.so.0
#30 0x00007f9eefbbbddd in ucp_init (context_p=, config=, params=0x7ffd30580310) at /usr/include/ucp/api/ucp.h:1913
#31 MPIDI_UCX_mpi_init_hook (rank=rank@entry=0, size=size@entry=1, appnum=appnum@entry=0, tag_bits=tag_bits@entry=0x7ffd30580418, init_comm=init_comm@entry=0x7f9ef0545b60 <MPIR_Comm_direct>) at src/mpid/ch4/netmod/ucx/ucx_init.c:241
#32 0x00007f9eefb9ad6f in MPID_Init (requested=, provided=provided@entry=0x7f9ef054dda0 <MPIR_ThreadInfo>) at src/mpid/ch4/src/ch4_init.c:595
#33 0x00007f9eefa69679 in MPIR_Init_thread (argc=argc@entry=0x7ffd3058093c, argv=argv@entry=0x7ffd30580930, user_required=, provided=provided@entry=0x7ffd305808ac) at src/mpi/init/initthread.c:158
#34 0x00007f9eefa69444 in PMPI_Init (argc=0x7ffd3058093c, argv=0x7ffd30580930) at src/mpi/init/init.c:131
#35 0x0000000000401c62 in ?? ()
#36 0x00007f9eef47e34d in __libc_start_main () from /lib64/libc.so.6
#37 0x000000000040236a in ?? ()