5.4-p1 crash

[Philippe PEGON]

Robert Watson a écrit :
> This sounds very similar to a serial console related tty bug I was 
> experiencing on -STABLE a few months ago, and that is believed may have 
> been worked around in 5.4 tweaks before release.  In particular, that 
> there are reference counting related bugs in the 5.x tty code that are 
> fixed by a partial rewrite of the tty code in 6.x, but that are too 
> large and disruptive to merge to RELENG_5.  If the problem is 
> persisting, it may be worth trying to merge anyway, but it is a pretty 
> big change and would break device driver binary compatibility, etc.  
> What we might want to do here is wait until 6.x has settled out a bit 
> more, then consider merging it to 5.x once 6.x has gotten burned in with 
> similar workloads and continued to not illustrate the 5.x tty reference 
> bugs.

Thanks for your answer.
Like I said on anothers posts, we have a FreeBSD 5.4-p1 which connects every fifteen minutes with an 
expect program to a lot of network devices for retrieving some informations, it seems that it is the 
culprit, the server crashed almost everyday. We reduced the frequency to one per hour and that 
attenuates the problem.
This panic is easy to reproduce with this simple expect program (see below) by running it 6 times 
simultaneously and waiting a few hours, I tested it on a HP DL360 with 2 cpu. If that can help, I 
can test this on current next week.


#! /usr/local/bin/expect

set timeout 60
set host [lindex $argv 0]

set pass "PASSWORD"

spawn ssh user at another-server

expect {
   "continue*(yes/no)" { send "yes\r" ; exp_continue }
   "assword:" { send "$pass\r" }
}

expect "*# " {
   send "ls\r"
}
expect "*#" {
   send "exit\r"
}

puts "Done."

> 
> Robert N M Watson

--
Philippe PEGON