[Top] [All Lists]

Re: oops on umount

To: Jens Axboe <axboe@xxxxxxx>
Subject: Re: oops on umount
From: Steve Lord <lord@xxxxxxx>
Date: Fri, 31 Mar 2000 11:34:42 -0600
Cc: lord@xxxxxxx, linux-xfs@xxxxxxxxxxx
In-reply-to: Your message of "Fri, 31 Mar 2000 17:29:20 +0200
Sender: owner-linux-xfs@xxxxxxxxxxx
> On Thu, Mar 30 2000, lord@xxxxxxx wrote:
> > > Looks like it tried to grab a spin lock in use and got caught by
> > > the nmi watchdog.
> > 
> > Hmm, this one could be fun - I suspect we have one thread using a buffer
> Indeed.
> > while another one is attempting to free it.... I have definitely seen
> > problems in this area. CONFIG_PAGE_BUF_META is going to be a fairly
> > delicate area for a while, we have implemented a totally new buffering
> > system underneath XFS - and it does not have exactly the same behavior
> > as the original irix one.
> Okay, this brings me to another point - what is your preferred bug
> report style? kdb or ksymoops? Should the PAGE_BUF_META options
> be enabled? Etc.

PAGE_BUF_META is the long term direction, but is only for the adventurous
(or developers) right now. It is a brand new buffer interface and has some
holes as far as XFS is concerned. So maybe in a few days I can shake this
out some more and turn off the old syle buffers.

kdb is the preferred debug tool - see other email on the list in the
last couple of hours. Supposedly an updated kdb is on its way in the
next day or so. Once people learn the code a bit the kdb modules for
pagebuf and xfs can give us formatted dumps of complex data structures.


<Prev in Thread] Current Thread [Next in Thread>