Subject: Re: another lfs deadlock, this time with consequence
To: None <tls@rek.tjls.com>
From: Blair Sadewitz <blair.sadewitz@gmail.com>
List: current-users
Date: 11/29/2006 08:59:07
No, I actually mean deadlock; the system itself was responsive, just
filesystems were broken.  Keyboard LEDs would turn on and off, I could
switch VTs, etc.  The first two I had were stuck in biowait, one was
stuck in vget (I think), and this most recent one I do not know.

It also occurred to me a while back that I was using BUFQ_PRIOCSCAN at
the time of the ones that were stuck in biowait.  I was *NOT* using
BUFQ_PRIOSCAN for the one stuck in vget or the most recent one.

I did not have DDB compiled into the kernel at the time. :(  But if I
do next time, is there anything specific I should ask it for?  I'm
unfamiliar with doing useful things with DDB on a whole.  I think I'm
going to try to rig up this laptop I have so I can use KGDB with it on
a serial console, or maybe just a plain serial console.

So, do you mean that fsck is superfluous, or that it has some value
aside from the roll-forward agent, which will likely be deleterious?
I think before 4.0 that option you're talking about should be added
and encouraged, then, unless the problem with roll-forward are fixed.



On 11/28/06, Thor Lancelot Simon <tls@rek.tjls.com> wrote:
> On Tue, Nov 28, 2006 at 01:14:26AM -0500, Blair Sadewitz wrote:
> >
> > I had another LFS deadlock tonight, but this time I couldn't even run
> > 'ps' or anything to see what was going on.
>
> Do you actually mean "deadlock", or do you just mean that your system
> froze?
>
> Did you try to enter DDB?
>
> > Upon reboot, fsck failed
> > with an unexpected inconsistency.  Well, the unexpected inconsistency
> > is that fsck sefaults after phase 6 has completed.  I can mount the
> > filesystem and it looks OK, and fsck didn't pick up any errors, but I
> > wonder how long this will go on for ...
>
> It is not necessary to run fsck on LFS filesystems at boot time, no
> matter how the system was restarted.  In fact, since there's at least
> one undiagnosed bug in the roll-forward agent, it may not even be
> advisable!
>
> I keep meaning to add an option to fsck_lfs to run everything _but_
> pass 6.
>
> Thor
>


-- 
Support WFMU-FM: free-form radio for the masses!

<http://www.wfmu.org/>
91.1 FM Jersey City, NJ
90.1 FM Mt. Hope, NY

"The Reggae Schoolroom":
<http://www.wfmu.org/playlists/RS/>