Firefox/Thunderbird and SSL not working with libthread_xu
dillon at apollo.backplane.com
Sat Apr 12 10:47:06 PDT 2008
:> :strange. Below is a commented extract from the 'kdump -T' output.
:> :// thread #4 calls umtx_sleep on 0x2a92f664
:> :// thread #4 never returns from this syscall, according to the kdump output
:> :75875:4 firefox-bin 1207996217.713757 CALL umtx_sleep(0x2a92f664,0,0)
:> :// thread #1 attempts to wake up thread #4
:> :75875:1 firefox-bin 1207996244.646643 CALL
:> :// I terminate the process
:> :75875:5 firefox-bin 1207996413.782573 PSIG SIGQUIT SIG_DFL
:> :The expected behaviour is for thread #4 to wake up when thread #1 calls
:> :I'd appreciate any feedback on the issue.
:> :If you need information about my environment, I'd be happy to provide it.
:> Could you paste all the ofktrace output between those two points? It
:> definitely should have woken up thread 4.
:> Matthew Dillon
:> <dillon at backplane.com>
:There are about 2 million lines of kdump output between these two
:points, which include the output of some debugging printfs and likely
:information read from/written to my personal profile. The only mentions
:of 0x2a92f664 in the kdump output are the lines pasted above. And, as I
:said, no return from umtx_sleep for thread 4 is noticed by ktrace.
:Before I terminated the process I ran 'ps -axHl' which resulted in the
:following line, among others:
: UID PID TID PPID CPU PRI NI VSZ RSS WCHAN STAT TT
: 1001 75875 4 74331 79 200 0 54880 43764 umtxsl IL+ p6
:Notice the 'I' in 'IL+'.
:The issue is very easy to reproduce - just run 'ktrace -i firefox' and
:try to open an SSL site. If opening the site fails, then close the
:window, wait for some time, then terminate the process. If opening the
:SSL site succeeds, just try again.
:I've reproduced the issue with thunderbird too, where the thread than
:never woke up happened to be thread #5. I'm running DragonFly 1.12.1
:with UP kernel.
:I ask anybody who doesn't trust my results to try and reproduce the issue.
I trust them, I just want to see what else is going on... for example,
if the memory in question is remapped inbetween the sleep and the
2 million lines isn't too bad, but not something to post to a mailing
list. Could you throw them up on a machine that I can download from?
I can give you a leaf account if you need a machine to upload the
<dillon at backplane.com>
More information about the Bugs