2006-10-30 12:14:22

by Metathronius Galabant

[permalink] [raw]
Subject: user-space command "ipcs" seems broken on 2.6.18.1

Hi,

the user-space command "ipcs" seems somewhat broken on my 2.6.18.1 as
it doesn't output anything, uses 100% CPU and isn't kill'able.
strace'ing it shows nothing/simply blocks.

Distro is CentOS 4.4, Kernel 2.6.18.1 x86 SMP, gcc version 3.4.6
20060404 (Red Hat 3.4.6-3).
Google didn't turn up anything. Any suggestions to track that down?

Cheers,
M


2006-10-30 12:32:14

by Jesper Juhl

[permalink] [raw]
Subject: Re: user-space command "ipcs" seems broken on 2.6.18.1

On 30/10/06, Metathronius Galabant <[email protected]> wrote:
> Hi,
>
> the user-space command "ipcs" seems somewhat broken on my 2.6.18.1 as
> it doesn't output anything, uses 100% CPU and isn't kill'able.
> strace'ing it shows nothing/simply blocks.
>
> Distro is CentOS 4.4, Kernel 2.6.18.1 x86 SMP, gcc version 3.4.6
> 20060404 (Red Hat 3.4.6-3).
> Google didn't turn up anything. Any suggestions to track that down?
>
Can you identify the latest kernel where it works OK?

--
Jesper Juhl <[email protected]>
Don't top-post http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please http://www.expita.com/nomime.html

2006-10-30 13:15:43

by Metathronius Galabant

[permalink] [raw]
Subject: Re: user-space command "ipcs" seems broken on 2.6.18.1

> Can you identify the latest kernel where it works OK?

I can't reproduce that behaviour on another SMP machine with the same
kernel-config for 2.6.18.1 (only storage and network device drivers
differ).
The affected one is a production machine I can't use to test, and
furthermore the only one I've got of that series.

Has to wait until the weekend. Any remote clue so I know where to look?
Thanks,
M

2006-10-30 17:33:28

by Jesper Juhl

[permalink] [raw]
Subject: Re: user-space command "ipcs" seems broken on 2.6.18.1

On 30/10/06, Metathronius Galabant <[email protected]> wrote:
> > Can you identify the latest kernel where it works OK?
>
> I can't reproduce that behaviour on another SMP machine with the same
> kernel-config for 2.6.18.1 (only storage and network device drivers
> differ).
> The affected one is a production machine I can't use to test, and
> furthermore the only one I've got of that series.
>
Ok, can you then at least tell us what the latest kernel you have used
that was OK was?
Just to try and narrow things down a bit.

> Has to wait until the weekend. Any remote clue so I know where to look?

Well, with a simple good/bad test case like you have, the obvious
thing to do would be to find a resonably new kernel that's good and
one that's bad and then do a git bisection search to find the exact
commit that broke things for you. Short of that, narrowing it down to
a released version, a -rc or -git snapshot is also good.

You could also start browsing through changelogs looking for changes
to IPC and then try revert patches that look likely. But git bisect
is probably a lot easier.

--
Jesper Juhl <[email protected]>
Don't top-post http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please http://www.expita.com/nomime.html

2006-10-30 17:36:13

by Jesper Juhl

[permalink] [raw]
Subject: Re: user-space command "ipcs" seems broken on 2.6.18.1

On 30/10/06, Jesper Juhl <[email protected]> wrote:
> On 30/10/06, Metathronius Galabant <[email protected]> wrote:
> > > Can you identify the latest kernel where it works OK?
> >
> > I can't reproduce that behaviour on another SMP machine with the same
> > kernel-config for 2.6.18.1 (only storage and network device drivers
> > differ).
> > The affected one is a production machine I can't use to test, and
> > furthermore the only one I've got of that series.
> >
> Ok, can you then at least tell us what the latest kernel you have used
> that was OK was?
> Just to try and narrow things down a bit.
>
> > Has to wait until the weekend. Any remote clue so I know where to look?
>
> Well, with a simple good/bad test case like you have, the obvious
> thing to do would be to find a resonably new kernel that's good and
> one that's bad and then do a git bisection search to find the exact
> commit that broke things for you. Short of that, narrowing it down to
> a released version, a -rc or -git snapshot is also good.
>
> You could also start browsing through changelogs looking for changes
> to IPC and then try revert patches that look likely. But git bisect
> is probably a lot easier.
>
Ohh and btw, testing the latest -rc (and/or -git snapshot) kernel is
also useful, to see if the bug has already been fixed in what is to
become 2.6.19 eventually.

--
Jesper Juhl <[email protected]>
Don't top-post http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please http://www.expita.com/nomime.html