On 11/15/06, Vivek Goyal <[EMAIL PROTECTED]> wrote:
> On Tue, Nov 14, 2006 at 12:49:17PM +0900, Magnus Damm wrote:
> > >> Current status of kexec-testing tree:
> > >>
> > >> Kexec - working well.
> > >> Kdump - 2.6.18 is broken on x86_64.
> > >
> > >I just tested kdump on 2.6.18.2 with horms's kexec-tools-testing tree
> > >and it is working fine. Actually in 2.6.19-rc kernels kdump is broken
> > >because of early param changes and your patch fixes it. But early
> > >param changes were not even present in 2.6.18 kernels so it should not
> > >be broken because of that.
> >
> > I agree with the early param stuff for 2.6.19-rc, but when you tested
> > 2.6.18,did you just kexec-on-panic or did you also try transfer the
> > vmcore?
> >
>
> I even transferred the vmcore and opened it using gdb and crash. Everything
> looked fine.
>
> We have been using 2.6.18 base kernel in RHEL, and kdump is working fine.

Are you using the kexec-tools-testing tree, or Eric's kexec-tools git tree?

> > >At the same time, in your last post you reported 2.6.18 kdump a success
> > >with -kdump10 kexec-tools. I think just change of version of kexec-tools
> > >should not break it.
> >
> > No, it _should_ not break it, but I think we have a regression in
> > kexec-tools-testing on x86_64 with older kernels.
> >
>
> Strange.

Indeed.

> [..]
> > >
> > >Ok. This warning message will pop up whenever latest kexec-tools is used
> > >with kernel older than 2.6.19-rc kernels. Now we parse /proc/kcore to
> > >find out where the kernel is loaded in virt addr space. Previous kernels
> > >needed a /proc/kcore fix and that fix is available only in 2.6.19-rc
> > >kernels.
> > >
> > >So above is just a warning message and does not harm. Please suggest if
> > >there is a better way to handle the things.
> >
> > Are you sure that they do no harm?
> >
>
> Yes, until and unless somebody is using old kernels where by default
> kernel was being compiled for address 1MB. There one shall have to use
> older version of kexec-tools with -kdump10.

Well, my config has CONFIG_PHYSICAL_START=0x100000, but I changed it
to 2 MB and it does not seem to help - still not able to read out
/proc/vmcore correctly.

While at it, was there any specific reason to not support 1MB physical
start address? It does seem to be a separate issue to my original
problem, but I think supporting any kind of sane CONFIG_PHYS_START
value sounds like a good thing to do IMO.

> > I'm not sure why we want to warn to the user about a missing feature.
> > Isn't it instead possible to just omit the PT_LOAD header that maps
> > the kernel if we cannot determine the virtual address? That may just
> > complicate things though...
> >
> > Also, if we really want to try hard then it should be possible to
> > fallback on information from /proc/kallsyms if present.
> >
> > >> ** cat /proc/vmcore fails:
> > >> / # cat /proc/vmcore | ftpput -u ftp -p ftp
> > >> 172.17.106.1 /incoming/vmcore -
> > >> cat: Read Error: Bad address
> >
> > The problem above is my main concern. Kexec-on-panic seems to work
> > properly, but that is kind of pointless if we cannot read out data
> > from /proc/vmcore...
> >
> > I've attached the config for my primary 2.6.18-kernel. It is kind of
> > minimalistic because I boot over the network and use initramfs instead
> > of root filesystem. The configuration works with the kexec-tools tree
> > but /proc/vmcore transfer is broken with kexec-tools-testing on
> > x86_64.
> >
>
> I will test with your config as soon as I am done with more modification
>  required for x86_64 relocatable kernel patches.

Cool. I'd like to try out your patches, but right now I can't kexec
x86_64 using kexec-tools-testing and 2.6.18, and I feel like tracking
it down.

So I will spend some time on this. If you find a working x86_64 config
for 2.6.18 please send it to me.

> So the bottom line is that with your config, you can't save vmcore on
> your system with 2.6.18 kernels and right now nature of problem is not
> known and no fix is available?

Correct.Using 2 MB instead of 1 MB does not seem to work either.

I'm using two 2.6.18 kernels - one primary and one secondary. Using
these two kernels I can successfully kexec-on-panic and transfer the
vmcore using code from the kexec-tools git repository. The latest
kexec-tools-testing git version does not work:

PASS: kexec-tools 529ad18980e297efc6ac2839c82afc24eccdcd1f
FAIL: kexec-tools-testing 785f8f7c4e71803cc1ec73f0660e28e14696d1cf

Thanks,

/ magnus
_______________________________________________
fastboot mailing list
[email protected]
https://lists.osdl.org/mailman/listinfo/fastboot

Reply via email to