Re: kern/38927: processes getting stuck in uvm_map (cv_timedwait), hanging machine

[Andrew Doran <ad%netbsd.org@localhost>]

The following reply was made to PR kern/38927; it has been noted by GNATS.

From: Andrew Doran <ad%netbsd.org@localhost>
To: Geoff Wing <gcw%pobox.com@localhost>
Cc: gnats-bugs%NetBSD.org@localhost
Subject: Re: kern/38927: processes getting stuck in uvm_map (cv_timedwait), 
hanging machine
Date: Thu, 12 Jun 2008 12:22:18 +0000

 On Thu, Jun 12, 2008 at 08:18:56PM +1000, Geoff Wing wrote:
 
 > On Wednesday 2008-06-11 11:49 +0100, Andrew Doran output:
 > :On Wed, Jun 11, 2008 at 04:16:35PM +1000, Geoff Wing wrote:
 > :>   nentries = 3220, 
 > :>   size = 999698432, 
 > :>   ref_count = 1, 
 > :It looks like something has exhausted the virtual address space available
 > :to the kernel (1GB). What is the machine doing?
 > 
 > sh /etc/security
 > 
 > It's doing a "find" for setuid/gid files.
 > 
 > If this is relevant, I have kern.maxvnodes=1048576
 > and "pstat -T" shows it make vnode use to around 532686
 
 That's why your machine is hanging: the kernel address space is so full that
 the system can no longer map anything into it. It should really do a better
 job of reclaiming space but that's difficult, and perhaps for now we should
 just limit the number of vnodes based on the available space.
 
 Can you 'print uvmexp' if you have a moment?
 
 Andrew