kern/104406: [ufs] Processes get stuck in "ufs" stateunderpersistent CPU load

[Kostik Belousov]

The following reply was made to PR kern/104406; it has been noted by GNATS.

From: Kostik Belousov <kostikbel at gmail.com>
To: bug-followup at FreeBSD.org
Cc: Oleg Derevenetz <oleg at vsi.ru>
Subject: Re: kern/104406: [ufs] Processes get stuck in "ufs" stateunderpersistent CPU load
Date: Tue, 23 Oct 2007 12:54:15 +0300

 --wIc/V6YLA2QdyfT4
 Content-Type: text/plain; charset=us-ascii
 Content-Disposition: inline
 Content-Transfer-Encoding: quoted-printable
 
 On Mon, Oct 22, 2007 at 04:29:32PM +0400, Oleg Derevenetz wrote:
 > I posted a followup to kern/104406 that includes all information listed i=
 n=20
 > "Debugging Deadlocks" chapter of FreeBSD Developer's Handbook. Can anyone=
 =20
 > take a look on it and say - is this certainly a hardware problem or some=
 =20
 > sort of software problem ?
 
 The ps output from ddb would simplify the analysis.
 
 The buffer space on the machine seems to be contaminated with a critical
 amount of the dirty buffers that are sent for the write. The pid
 844 (that is the only one producing the writes) is blocked in
 waitrunningbufspace(), waiting for the write queue to be drained. There
 are other processes, for instance, 937 etc, that wait for the buffer
 lock. Most likely, these buffers are scheduled for write and cause the
 processes to be uninterruptibly blocked.
 
 If this state of the system is static, then the write queue is not
 processed. Most plausible cause is the controller (mly0) or the controller
 driver fault.
 
 --wIc/V6YLA2QdyfT4
 Content-Type: application/pgp-signature
 Content-Disposition: inline
 
 -----BEGIN PGP SIGNATURE-----
 Version: GnuPG v2.0.4 (FreeBSD)
 
 iD8DBQFHHcTGC3+MBN1Mb4gRAodzAKDLMJIiPnjWNRejDrnj/6R/Kic1LwCbBxK+
 9GLcje2brT6bTZA1dzB17Yg=
 =2chV
 -----END PGP SIGNATURE-----
 
 --wIc/V6YLA2QdyfT4--