On Tue, Mar 09, 2010 at 12:47:01PM -0500, Thorsten Brehm wrote:
> Hi,
> we???re (again) having an issue with io-pkt crashing when disconnecting ppp connections (e.g. start multiple pppds + ?
?? slay pppd???). Target is x86.
> With our original support case (CaseID00100180) we could reproduce the problem with at least 3  ppp connections. 
Shutting down all three connections would always cause io-pkt-v4 to crash (NULL pointer exception).
> 
> We closed the support case (and I guess PR#73053) since we could no longer reproduce the problem with revision 1139 
(io-pkt main trunk, fixing PR71330 + PR71018). We now noticed the problem still exists ??? but it takes at least 6 pppd 
connections to crash io-pkt now. The crash always happens at the same moment, e.g. when shutting down the final 
connection ??? so it???s obviously not a rare timing issue.
> 
> We have also tested the latest revision from the 6.4.1 branch (io-pkt revision 149) ??? same result.
> 
> When using a debug build we see an assertion failing:
> In function receive_loop_uni  Z:/io-pkt-rev149/6.4.1/trunk/sys/receive.c:1012 ((wpt))->wt_critical == 0 -- assertion 
failed
> 
> We started to dig a little deeper and found the assertion is triggered by two consecutive calls to ???NW_SIGUNHOLD_P??
? (which causes wt_critical == -1).
> We logged the calls to ???NW_SIGHOLD_P??? and ???NW_SIGUNHOLD??? and get:
> 
> NW_SIGHOLD_P:   wpt->wt_critical=0  wpt: 0x80e7b00, m_ext_free:830
> NW_SIGUNHOLD_P: wpt->wt_critical=1  wpt: 0x80e7b00, pool_cache_put_header:2764
> NW_SIGHOLD_P:   wpt->wt_critical=0  wpt: 0x80e7b00, m_ext_free:830
> NW_SIGUNHOLD_P: wpt->wt_critical=1  wpt: 0x80e7b00, pool_cache_put_header:2764
> NW_SIGHOLD_P:   wpt->wt_critical=0  wpt: 0x80e7b00, m_ext_free:830
> NW_SIGUNHOLD_P: wpt->wt_critical=1  wpt: 0x80e7b00, pool_cache_put_header:2764
> NW_SIGUNHOLD_P: wpt->wt_critical=0  wpt: 0x80e7b00, pool_cache_get_header:2627
> NW_SIGHOLD_P:   wpt->wt_critical=-1 wpt: 0x80e7b00, m_ext_free:830
> NW_SIGUNHOLD_P: wpt->wt_critical=0  wpt: 0x80e7b00, pool_cache_put_header:2764
> NW_SIGHOLD_P:   wpt->wt_critical=-1 wpt: 0x80e7b00, process_pkts:265
> NW_SIGUNHOLD_P: wpt->wt_critical=0  wpt: 0x80e7b00, receive_loop_uni:1012
> Z:/io-pkt-rev149/6.4.1/trunk/sys/receive.c:1012 ((wpt))->wt_critical == 0 -- assertion failed
> 
> Line numbers may be off by 1-2 lines (compared to rev 149) since we added some lines for logging.
> 
> Before we start spending any time to dig even deeper ??? any ideas/hints on the issue?

Yes, that particular count is off; however, I don't think
that in itself would cause an issue other than the assert
in the debug build as the underlying lock looks to be
handled in a consistent manner.  At any rate I've made a PR
on this lock issue.

Regards,

-seanb