Re: ax25_recvmsg lockup

[Dave Jones <davej@redhat.com>]

On Sat, Aug 03, 2013 at 01:46:04PM -0700, David Miller wrote:
 > From: Dave Jones <davej@redhat.com>
 > Date: Mon, 17 Jun 2013 10:28:03 -0400
 > 
 > > With the right arguments from userspace, reading from an ax25 socket
 > > can spend long enough in the kernel that it triggers the softlockup
 > > detector.  I hit a bunch of these traces in a trinity run that I
 > > left running over the weekend.
 > 
 > I looked at this a bunch of times and am still mystified how this
 > can happen.
 > 
 > It should be simple to diagnose since I doubt there is any real
 > AX25 traffic going to these sockets.
 > 
 > Any chance you can figure out what the arguments are that cause
 > this problem?  That'll probably help a lot, thanks!

thinking about this some more, I wonder if it was related to the problems
I was seeing where a bunch of concurrent sync()'s would trip up the
soft-lockup detector.  Perhaps the ax25 code got preempted and
the syncs starved the cpu for a while, so when we eventually got back
to the ax25 code, the detector decided it hadn't made progress..

I don't know, it sounds a bit far-fetched, but it's all I can think of.

	Dave