kexec-tools question

[larry.brigman at gmail.com (Larry Brigman)]

On 12/21/06, Magnus Damm <magnus.damm at gmail.com> wrote:
> On 12/22/06, Larry Brigman <larry.brigman at gmail.com> wrote:
> > Trying to debug a problem and my understanding is dim.
> >
> > Is the entry that is presented to the kernel equal to
> > the entry that one would get from readelf -h on a vmlinux compiled with
> > a physical offset?
>
> On a vmlinux (2.6.19 to be more exact):
>
> So you have the startup_32 symbol in the kernel that you want to boot.
> It should be in arch/i386/kernel/head.S.
>
> Then you have the linker script in arch/i386/kernel.vmlinux.lds.S
> which sets ENTRY() to phys_startup_32 which is startup_32 -
> PAGE_OFFSET ie the physical address of startup_32.
>
> Linux-2.6.19 is known to work to boot into on i386. Use that as a
> secondary kernel to begin with and when you have that working you can
> try booting older kernels. I have a small working serial console
> config if you are interested.
>
> The kernel entry point is not everything though. You have purgatory
> which is a glue layer in between the kernels. It is contained in the
> kexec-tool and loaded and linked during runtime.
>
> So the entry point you will see in kexec (in your first kernel) will
> most likely be the entry point of purgatory. It took me a while to
> figure that out and now I have mixed feelings about purgatory... =)
>
> / magnus
>
What should the i386 entry value be on the kexec_load call on i386
if the CONFIG_PHY_START = 0x1000000 for the crash dump kernel

Without debug on I get 0x1498 with caused the kernel to reject the
load.

I think I am finding a silent failure in the kexec code.  I turned on
debug in the kexec code and get a different error.
unable to open /sys/devices/system/cpu/cpu0/crash_notes

I think that this is one of two things.  I don't have crash dump turned
on in the first kernel or I missed the section in the sysfs that adds
thje crash_notes section.
Have to wait until tomorrow to figure it out.