lkcd
[Top] [All Lists]

Re: Problems with 2.4.4 crash dumps

To: <lkcd@xxxxxxxxxxx>
Subject: Re: Problems with 2.4.4 crash dumps
From: "Matt D. Robinson" <yakker@xxxxxxxxxxx>
Date: Mon, 4 Jun 2001 00:36:32 -0700 (PDT)
Cc: Matt Robinson <yakker@xxxxxxxxxxx>, <simon.falvey@xxxxxxxxxxx>
Sender: owner-lkcd@xxxxxxxxxxx
Simon Falvey wrote:
> Hi.
>
> This may be a bit of a newbie question. I have only just discovered lkcd and
> I think I am going to get a lot of value out of it in my job (surely not?
> work related linux who ever heard of such a thing).

If you're in support, it can be useful ... :)

> I applied by hand the patch for 2.4.1 to the 2.4.4 kernel running under
> RH7.1. I have also placed a hook under the Magic sysrq key sequence hook a
> call to panic() on demand (This will be very useful for diagnosing hung
> systems).
>
> The system falls into the dump_execute  routine which then calls
> alloc_kiovec in the section "start walking through the page tables ". The
> system then does not return (to dump_execute)  from this. On the screen it
> reports a BUG in slab.c line 1073 which is a call to BUG(). It then returns
> back to panic() (I think)  without finishing the dump but it does finish the
> panic and reboot the system. Such that..
>
>                         /* debug: print markers to trace the problem */
>                         printk("Marker 1 ");
>                 /* start walking through the page tables */
>                 if (alloc_kiovec(1, &dump_iobuf)) {
>                                 printk("Marker 2 ");
>                         printk("\n" KERN_WARNING
>                                 "alloc_kiovec() failed!");
>                 } else {
>                 printk("Marker 4 ");
>
> Neither Marker 2 nor Marker 4 are reached.

Two things changed - the alloc_kiovec() functionality changed
between 2.4.1 and 2.4.4.  Since I've been porting the patches
forward, I'm going to move the alloc_kiovec() code up into the
dump_open() function so this problem doesn't occur again.  That
way you don't have to worry about the changes to the kiovec code.

> I am entirely satisfied that the patch has applied correctly. The
> configuration of vmdump is a level 4 with compression to /dev/hda6 (the swap
> dev).
>
> Has anything similar been seen before? Has any one any ideas about the
> problem?
> Also any idea when the lkcd code will make the standard linux code base?

I haven't seen it before, but then again, I haven't tried this on
2.4.4 (yet).  I've built a 2.4.2 patch, and I'm building a kernel
now.  If this patch works, I'll put these on oss.sgi.com now.

> Thanks
>
> Simon

I hope this helps.  More later tonight ...

> Simon Falvey
> Online Product Support Specialist
> VERITAS Software UK.
> Tel: +44 118  918 8105

--Matt


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