Re: Should we drop non-SMP ia64 kernel images?
On Tue, 2005-09-27 at 13:38 -0600, dann frazier wrote:
> On Tue, 2005-09-27 at 11:00 -0700, Christoph Lameter wrote:
> > On Tue, 27 Sep 2005, Sven Luther wrote:
> > > Do you mean that kexec doesn't work in SMP situations ? I am not sure what you
> > > mean about test situations and the SMP kernel should work just well in
> > > virtualized environments, but i have some trouble equating ia64 machines with
> > > low memory situations :)
> > The new kernel started with kexec for debugging purposes may need to be
> > simple and small. AFAIK the amount of memory for the second kernel is
> > limited. Virtualization of SMP environments is pretty complex. In
> > many situations its just nice to have a UP kernel around.
> Khalid: I know you've been hacking on kexec a lot lately - can you
> comment on the kexec case? Do you know if such a limitation exists, and
> what it is?
I have not come across any such limitation. I kexec a full blown SMP
kernel with lots of things built-in a few times every day. The only
issue I can think of is that currently running kernel has to load a to
be kexec'd kernel in memory and it kmalloc's memory to hold this image.
If kernel does not have enough memory to load a new kernel, then you
could see a limitation on how big a kernel can be kexec'd, but that
situation is no different from running the system out of memory and not
being able to kmalloc memory for any other reason as well.
> Christoph: What other cases might you think of where booting an SMP
> kernel with maxcpus=1 is not sufficient?
> The only concern I had prior to starting this thread is the performance
> overhead; I'm interested in any benchmarks that might be able to
> demonstrate a difference here.