[Top] [All Lists]

Re: corruption on 2.4.6 w/ low memory?

To: knuffie@xxxxxxxxx (Seth Mos)
Subject: Re: corruption on 2.4.6 w/ low memory?
From: Sidik Isani <isani@xxxxxxxxxxxxxxx>
Date: Tue, 2 Oct 2001 12:22:13 -1000 (HST)
Cc: isani@xxxxxxxxxxxxxxx (Sidik Isani), linux-xfs@xxxxxxxxxxx
In-reply-to: <> from "Seth Mos" at Oct 02, 2001 10:07:51 PM
Sender: owner-linux-xfs@xxxxxxxxxxx
<Seth Mos wrote:>
|At 08:19 2-10-2001 -1000, Sidik Isani wrote:
|>Hello -
|>   I'm having a problem with XFS and/or the 2.4.6 kernel.  Since
|>   2.4.6+XFS is not an official release, I'm guessing the first
|>   thing would be to use 2.4.5-xfs-1.0.1... unless this is a known
|>   problem which 2.4.5 might have as well?  In any case, I'd like
|>   your advice on which version to use, to try and recreate this
|>   and get more debugging information.  (I know what I've included
|>   is probably not enough.)
|Can you try a CVS kernel?

  Sure.  Though I am testing this stuff for eventual use on
  production systems...

|>   The symptoms are files which were written *minutes* ago retain
|>   the right size, but seem to develop blocks full of zero bytes.
|>   I think this mostly happens when memory runs very low, but I'm
|>   not sure.  I'm running an SMP kernel, with no swap space, and I'm
|>   writing files to "tmpfs" at the same time.  (With the UP kernel,
|>   I've noticed a different, but possibly related problem when
|>   memory runs low where bdflush gets stuck taking 100% of the CPU.)
|What compiler did you use? Egcs 1.1.2 is the recommended compiler for 
|production use.

  gcc-2.95.3 -g -O

|>   Usually, there are no errors from the kernel while this is
|>   happening, but eventually I got these:
|>kernel BUG at ll_rw_blk.c:700!
|>invalid operand: 0000
|>   And "free" showed:
|>              total       used       free     shared    buffers     cached
|>Mem:        254120     243684      10436          0      51688     215380
|>-/+ buffers/cache:     -23384     277504
|>Swap:            0          0          0
|Why run without swap?

  On this machine, there is no good reason.  I will try to create some,
  and see if the same problem occurs when both swap and memory fill
  up or not.  Other similar machines need to be disk-less NFS-root
  but those won't exercise their XFS, code of course.  Is there a
  fundamental difference between having, say, 1 GB of RAM versus
  512MB RAM + 512 Swap?

Be seeing you,

- Sidik

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