Hi,
I would just like to add that I also see these hangs of unkillable
processes when trying to write large files to XFS. In my case it can take
up to half an hour until the process unhangs itself.
I am using SGI-supplied RedHat 8.0 kernel binaries from 1.2pre2
kernel-2.4.18-17SGI_XFS_1.2pre2.i686.rpm so this should make it
somewhat easier to reproduce this. The system I used is a
Dell Inspiron 4100 Notebook with 1.2 GHz P3 and 1 Gig of RAM.
AFAIK, RedHat 8 also uses a "small" version of the low latency patch,
so this could well be the problem.
Linux version 2.4.18-17SGI_XFS_1.2pre2 (root@xxxxxxxxxxxxxxxxxxxxxxx)
(gcc version 3.2 20020903 (Red Hat Linux 8.0 3.2-7)) #1 Tue Oct 22 22:38:47 CDT
2002
This is a real showstopper for Release 1.2, IMHO.
BTW, I'm actually a little desperate for finding a good stable kernel with
all features I need (XFS, IPsec, ALSA, 4 Gig, OpenMosix, VMware, DVB, ...).
Stock RH80 hangs on APM suspend, RH8-XFS hangs on large file write, Mandrake
9.0 does not support IDE DMA on ICH-3M and ICH-4 (besides not supporting
more than 896 MB RAM in the default config), Gentoo 2.4.19 has the same IDE
problem and is a little too heavily hacked for my taste since I still need
to add OpenMosix....
Cheers,
Dirk
------------------------------------------
Ingenieurbüro Dipl.-Ing. Dirk W. Steinberg
Email: dws@xxxxxxxxxxxxxxxx
Christian Lambert wrote:
>
> Eric,
>
> I can't reproduce it with the stock kernel with only supermount enabled so
> far.
> So maybe it's not playing well with the low latency patch even if just
> compiled in, but not enabled in /proc. Or it was because I was using
> the old xfs snapshot from 2.4.19 and not the 1.2pre-release. I'm gonna try
> to
> add them one by one and re-test.
>
> BTW, I think you should add the 1.2 pre-release in the snapshots on the
> webpage
> so that people don't download the older ones if 1.2pre is more recent and
> available. If I would have not checked the mailing list, I would have not
> found
> about 1.2pre. I looked at the weekly snapshot, and it hasn't been updated
> for a while now.
>
> -Christian
>
> On Tue, 29 Oct 2002, Eric Sandeen wrote:
>
> | Hi Christian -
> |
> | Reproducing this with the stock kernel & stock patches (preferably
> | the 1.2-pre patches) would be most helpful, isolating bugs with the
> | patch-set-du-jour can be rough.
> |
> | If you can't reproduce it with the stock kernel, add the patches
> | one at a time and see where things fall apart.
> |
> | -Eric
> |
> | On Tue, 29 Oct 2002, Christian Lambert wrote:
> |
> | >
> | > I now disabled pre-empt, and basically the only kernel patch
> | > I added was supermount+ O(1) batch scheduler and -aa vm patch. Now the
> problem
> | > seems to be different. If a process writes large files (about 1 or 2
> gigs)
> | > it will often hangs right when it reach the end of the file and it was
> | > about to update the directory entry, it just hangs there. Top shows 100%
> | > system usage (0% user) and I can't kill the process until it unhangs
> itself
> | > which takes about 2-3 minutes. I'm using the 2.4.19 snapshot, but I just
> | > saw that you have a 1.2 pre-release for 2.4.19, so I'm going to recompile
> | > my kernel with that and just supermount. I'm also using vmware (which has
> | > kernel modules) so i'm not sure if that has an impact or not, but I
> thought I'd
> | > mention it. I could reproduce it often but just doing a "cp -a dir1
> dir2"
> | > with a couple of 1 gig files in it.
|