[Crash-utility] strange kdump vmcore PT_LOAD segments

[Dave Anderson]

Vivek Goyal wrote:

> On Thu, May 04, 2006 at 03:08:14PM -0400, Dave Anderson wrote:
> > >
> > > Right, but something is fundamentally wrong with your
> > > vmcore:
> > >
> > >                   p_paddr  p_filesz
> > > First segment:          0      8000
> > > Second segment:      8000   1ff7fff
> > > Third segment:   2fd0f001  d02f0fff
> > >
> > > It cannot have a p_paddr address that is not page-aligned,
> > > nor can you have a p_filesz that is not page-aligned.
> > >
> > > Dave
> >
> > Vivek et al,
> >
> > Have there been any recent changes to the kdump code
> > that might cause strange PT_LOAD segment contents?
> >
>
> Dave, As per my knowledge nothing has changed. Generally PT_LOAD headers are
> generated from user space while kernel is being loaded and second kernel
> leaves them untouched except updating the offset field. So there are
> high chances that user space itself is not generating right headers while
> loading the capture kernel.
>
> Dave Wilder, can you put some debugging statement in kexec-tools and see
> what header values are being generated from user space and do they match
> with what /proc/vmcore is showing.
>
> I will do some experiments on x86_64.
>
> Thanks
> Vivek

OK thanks -- changing the $Subject to keep better track of this.

Dave