Re: [PATCH v5] x86, kdump: Set crashkernel_low automatically

[Vivek Goyal <vgoyal@redhat.com>]

On Wed, Mar 20, 2013 at 09:21:31AM -0700, Yinghai Lu wrote:
> On Wed, Mar 20, 2013 at 9:03 AM, Vivek Goyal <vgoyal@redhat.com> wrote:
> > On Wed, Mar 20, 2013 at 08:53:29AM -0700, Yinghai Lu wrote:
> >> On Wed, Mar 20, 2013 at 6:08 AM, Vivek Goyal <vgoyal@redhat.com> wrote:
> >>
> >> > Have a general question about crashkernel_low. Why does it need to
> >> > show up as "Crash kernel low" in /proc/iomem. Will it not be better
> >> > that all memory reserved for crashkernel (whether high or low), shows
> >> > as "Crash Kernel" and let kexec-tools decide whether to load kernel
> >> > high or low etc.
> >> >
> >> > IOW, there should not be any need to differentiate between "Crash kernel"
> >> > and "Crash kernel low". There are address ranges associated and looking
> >> > at addresses it is obivious that certain memory is below 4G.
> >>
> >> yes. it is doable.
> >> but
> >> 1. will need to add more code to expand parse_iomem_single to handle
> >> multiple "Crash kernel" in kexec-tools.
> >> 2. also we already have "crashkernel_low=" in command line, so it is
> >> good to keep them consistent in /proc/iomem.
> >
> > I think command line and /proc/iomem output are very different.
> > crashkernel_low is just enforcing that reserve it below 4G and memory
> > type still remains "Crash Kernel".
> >
> > So to me, /proc/iomem is showing ranges and memory type and both the
> > memory types should be "Crash Kernel".
> >
> > IMHO, we should add code in kexec-tools to deal with it (multiple
> > entries for memory type "Crash Kernel"), instead of especial casing
> > "Crash Kernel Low". Who knows down the line we end up reserving more
> > crash kernel memory which is not contiguous. Keeping all reserved
> > memory of same type will help then.
> 
> ok.
> 
> Need to fix kexec-tools at first, and the drop Low in kernel.
> 
> Before v3.9 and kexec-tools 2.0.4?

I think so. We need to do this in 3.9 otherwise it becomes another
backward compatibility issue.

Thanks
Vivek