|
Tomasz Lis
|
QNET module crashes when my .SO driver is already unloaded
|
|
Tomasz Lis
05/21/2010 2:15 AM
post55333
|
QNET module crashes when my .SO driver is already unloaded
To test my newly created TAP driver I made a few scripts, and I have a problem with this one:
---
#!/bin/sh
for a in 1 2 3 4 5 6 7 8 9 10 ... 100
do
ifconfig tap0 destroy
ifconfig || { echo "list interfaces failed"; exit 1; }
mount -T io-pkt -o mac=aaaaaaaaaaaa,verbose=7 devnp-rtap.so || { echo "mount failed"; exit 1; }
ifconfig tap0 10.0.0.1 netmask 255.255.255.0 || { echo "ifconfig failed"; exit 1; }
route add -host 10.0.0.2 10.0.0.1
done
---
It seems that sometimes after the .so file is unloaded, QNET is making segmentation fault (signal 11).
Even stranger is the fact that when I remove creation of a special thread I made for Resource Manager, this script seems
to work fine.
I've tried using both nw_thread and standard pthread; I also tried making 1s delay before my .so is unloaded (in detach
function), and using various functions from io-pkt (like quiesce_all() or nw_thread_untracked_add()). But nothing seems
to help - the library terminates,
usually on listing interfaces, after several tenths of iterations.
Here's a typical backtrace.
I don't have compilable source of io-pkt, so this is standard io-pkt-v4-hc from QNX 6.4.1:
---
#0 segv_handler (signo=11) at /home/builder/hudson/641/svn/lib/io-pkt/sys/main.c:498
498 *(int *)NULL = 0; /* segv for real now */
(gdb) #0 segv_handler (signo=11) at /home/builder/hudson/641/svn/lib/io-pkt/sys/main.c:498
No locals.
#1 <signal handler called>
No symbol table info available.
#2 0xb8220892 in try_ifp (ifp=0x81931c0)
at /home/builder/hudson/641/svn/lib/io-pkt/sys/lsm/qnet/l4_lite/l4_driver_en_iopkt.c:310
dp = <value optimized out>
ndb_if = <value optimized out>
nret = <value optimized out>
empty_slot = <value optimized out>
I = <value optimized out>
__FUNCTION__ = "try_ifp"
#3 0xb82298f4 in scavenger (ctp=0x8194778, code=0, flags=0, hdl=0x0)
at /home/builder/hudson/641/svn/lib/io-pkt/sys/lsm/qnet/ndb/ndb.c:268
chstate = 0
next = {it_value = {tv_sec = 2956070192, tv_nsec = 135521472}, it_interval = {tv_sec = 135874424,
tv_nsec = -1338782302}}
ifep = (struct ndb_ifent *) 0x813d620
ifepp = (struct ndb_ifent **) 0x813d740
iflist_tmp = (struct ndb_ifent *) 0x813d770
#4 0xb03349af in _message_handler (dctp=0x8194778)
at /home/builder/hudson/641/svn/lib/c/dispatch/message.c:329
i = 3
vec = (message_vec_t *) 0x813e490
def_vec = (message_vec_t *) 0x0
code = 0
mutex = (pthread_mutex_t *) 0x8140150
#5 0xb0333768 in dispatch_handler (ctp=0x81931c0)
at /home/builder/hudson/641/svn/lib/c/dispatch/dispatch.c:328
No locals.
#6 0xb8219188 in qdpp_handle_thread (arg=0x813f1d0)
at /home/builder/hudson/641/svn/lib/io-pkt/sys/lsm/qnet/l4_lite/iopkt.c:111
ctp0 = (dispatch_context_t *) 0x8194778
ctp = (dispatch_context_t *) 0x8194778
__FUNCTION__ = "qdpp_handle_thread"
#7 0x080c9d8b in thread_init (arg=0x813f1d0)
at /home/builder/hudson/641/svn/lib/io-pkt/sys/nw_thread.c:765
wtp = (struct nw_work_thread *) 0x8194e80
funcp = (void *(*)(void *)) 0xb82190b6 <qdpp_handle_thread>
err = <value optimized out>
#8 0xb031fa20 in ?? () from /usr/qnx641/target/qnx6/x86/lib/libc.so.3
---
How should I force qnet to leave my interface alone?
This is simplified code of my tap_detach() function:
---
tap_stop(ifp, 1);
if_down(ifp);
ether_ifdetach(ifp);
if_detach(ifp);
nw_pthread_reap(sc->rm.restid,sc);
interrupt_entry_remove(&sc->sc_inter, NULL);
shutdownhook_disestablish(sc->sc_sdhook);
---
And here's simplified cleanup function of the thread:
---
pulse_detach(sc->rm.dpp, sc->rm.quiesce_code, 0);
if (resmgr_detach(sc->rm.dpp, sc->rm.rid, 0) == -1) {
LOGE("%s: resmgr_detach failed: %s", DEV_NAME_TAP, strerror(errno));
}
sc->rm.rid = -1;
---
|
|
|
|
|
