Re: kernel bug: futex_wait hang

[Lee Revell <rlrevell@joe-job.com>]

On Mon, 2005-03-21 at 20:20 -0800, Andrew Morton wrote:
> Lee Revell <rlrevell@joe-job.com> wrote:
> >
> > Paul Davis and Chris Morgan have been chasing down a problem with
> > xmms_jack and it really looks like this bug, thought to have been fixed
> > in 2.6.10, is the culprit.
> > 
> > http://www.uwsg.iu.edu/hypermail/linux/kernel/0409.0/2044.html
> > 
> > (for more info google "futex_wait 2.6 hang")
> > 
> > It's simple to reproduce.  Run JACK and launch xmms with the JACK output
> > plugin.  Close XMMS.  The xmms process hangs.  Strace looks like this:
> > 
> > rlrevell@krustophenia:~$ strace -p 7935
> > Process 7935 attached - interrupt to quit
> > futex(0xb5341bf8, FUTEX_WAIT, 7939, NULL
> > 
> > Just like in the above bug report, if xmms is run with
> > LD_ASSUME_KERNEL=2.4.19, it works perfectly.
> > 
> > I have reproduced the bug with 2.6.12-rc1.
> > 
> 
> iirc we ended up deciding that the futex problems around that time were due
> to userspace problems (a version of libc).  But then, there's no discussion
> around Seto's patch and it didn't get applied.  So I don't know what
> happened to that work - it's all a bit mysterious.
> 

It does seem like it could be a different problem.  Maybe Paul can
provide some more evidence that it's a kernel and not a glibc/NPTL bug.
I'm really just posting this on Paul's behalf; I don't claim to
understand the issue. ;-)

> Is this a 100% repeatable hang, or is it some occasional race?
> 

100% repeatable.

Lee