guix/gnu/packages/patches/psm-disable-memory-stats.patch

63 lines
2.5 KiB
Diff
Raw Normal View History

The memory statistics code leads to segfaults during initialization (on
machines with InfiniPath networking):
(gdb) bt full
#0 ips_ptl_init (ep=0x1fc6af8, ptl=0x1fc6f88, ctl=0x1fc6d78) at ptl.c:224
err = PSM_OK
num_of_send_bufs = 1024
num_of_send_desc = 4096
imm_size = 128
context = 0x1fc6b70
user_info = 0x1fc6b90
enable_shcontexts = 0
current_count = <optimized out>
#1 0x00007fb2aa672abf in __psm_ep_open_internal (
unique_job_key=unique_job_key@entry=0x7ffed1ee5800 "<\207\020#5\271\267\200\354x\242e8\364zo",
devid_enabled=devid_enabled@entry=0x7ffed1ee5724, opts_i=opts_i@entry=0x7ffed1ee5810, mq=<optimized out>,
epo=epo@entry=0x7ffed1ee5710, epido=epido@entry=0x7ffed1ee5708) at psm_ep.c:929
ep = 0x1fc6af8
num_units = 1
len = <optimized out>
err = <optimized out>
epaddr = 0x1e9dd78
buf = "miriel044:2.0.", '\000' <repeats 113 times>
p = <optimized out>
e = <optimized out>
old_cpuaff = 0x0
old_unit = 0x0
yield_cnt = {e_void = 0xfa, e_str = 0xfa <error: Cannot access memory at address 0xfa>, e_int = 250,
e_uint = 250, e_long = 250, e_ulong = 250, e_ulonglong = 250}
no_cpuaff = {e_void = 0x0, e_str = 0x0, e_int = 0, e_uint = 0, e_long = 0, e_ulong = 0, e_ulonglong = 0}
env_unit_id = {e_void = 0xffffffffffffffff,
e_str = 0xffffffffffffffff <error: Cannot access memory at address 0xffffffffffffffff>, e_int = -1,
e_uint = 4294967295, e_long = -1, e_ulong = 18446744073709551615, e_ulonglong = 18446744073709551615}
env_port_id = {e_void = 0x0, e_str = 0x0, e_int = 0, e_uint = 0, e_long = 0, e_ulong = 0, e_ulonglong = 0}
env_sl = {e_void = 0x0, e_str = 0x0, e_int = 0, e_uint = 0, e_long = 0, e_ulong = 0, e_ulonglong = 0}
ptl_sizes = <optimized out>
default_cpuaff = <optimized out>
opts = {timeout = 180000000000, unit = -1, affinity = 0, shm_mbytes = 10, sendbufs_num = 1024,
network_pkey = 65535, port = 0, outsl = 0, service_id = 1152940698815692800,
path_res_type = PSM_PATH_RES_NONE, senddesc_num = 4096, imm_size = 128}
amsh_ptl = 0x1fc6e48
ips_ptl = 0x1fc6f88
self_ptl = 0x1fc99c8
i = 3
It looks like ptl.c:24 is writing past the region that was malloc'd.
Turning stats off solves the problem.
diff --git a/psm_utils.c b/psm_utils.c
index c8651fe..5514921 100644
--- a/psm_utils.c
+++ b/psm_utils.c
@@ -1058,7 +1058,7 @@ psmi_log_memstats(psmi_memtype_t type, int64_t nbytes)
return;
}
-#define psmi_stats_mask PSMI_STATSTYPE_MEMORY
+#define psmi_stats_mask 0
#ifdef malloc
#undef malloc