Re: [Bugme-new] [Bug 12201] New: long wait in call_usermodehelper() / queue_work() / wait_for_completion()

[Michael Spiegle <mike@nauticaltech.com>]

> It'd be great if you could test something more recent please.

Will do.  It takes a little bit of time to reproduce, so I'll respond 
now with information from my current kernel and update once the new 
kernel is running.

> Please attach a copy of the config to
> http://bugzilla.kernel.org/show_bug.cgi?id=12201

Done: http://bugzilla.kernel.org/attachment.cgi?id=19263

> Please get sysrq working then get us a task trace, so we can see who is
> sleeping where.  Do this:

Done: http://bugzilla.kernel.org/attachment.cgi?id=19264

I wanted to get the call to /sbin/modprobe in there as well, so I had to 
do it about 10 times before I could catch it.  I also stopped 
unimportant processes on that box to eliminate additional variables.

There is one other variable I didn't mention.  The machine is diskless 
and running on an nfs root.  I have tried to eliminate this variable via 
simple testing (just running /sbin/modprobe at the commandline returns 
instantly), but wanted to mention it.

The box was 100% idle before starting the pilot program and running the 
task trace.  I added gettimeofday() around my socket() call and got 
about 1290000us - 1490000us on average.  Using a socket() call with 
protocol NETLINK_ROUTE returns in about 5us - 8us.  Certain other 
protocol types like NETLINK_FIREWALL also have the same long delay.


Michael Spiegle
mike@nauticaltech.com