Re: [PATCH/RFC] Re: Kernel selection on i386 for Xen compatibility
On Wed, 2008-04-30 at 10:39 +0200, Frans Pop wrote:
> On Wednesday 30 April 2008, Ian Campbell wrote:
> > On Sun, 2008-04-27 at 12:30 +0100, Ian Campbell wrote:
> > > Shall I define a new environment variable ${DI_FLAVOUR} passed to the
> > > base-installer/kernel scripts containing the kernel suffix (i.e. -486
> > > vs -686-bigmem) or ...
> >
> > Something like this? (only tested with the tests in the tree so far not
> > a proper image)
>
> IIUC this would only work if the installer is already running a -bigmem
> kernel, which might be an option for custom versions of the installer, but
> IMO not for regular versions.
That was going to be the subject of my next post ;-)
A PAE-capable install image is required to support d-i on Xen (since the
majority of Xen installations are PAE). I don't know about other
hypervisor technologies particularly but I'd not be surprised if they
preferred PAE too since virtualisation tends to imply a >4G host machine
even if the individual VMs are smaller.
I'd really like it if Debian could supply a PAE enable installer as
standard but if not I'll find another way (presumably supplying one
myself).
I experimented with a netboot image with both -486 and -686-bigmem and
it added 1.4M to a 5M initrd.gz image, which I can see could be
unacceptable. A separate netboot-bigmem could be an alternative (note: I
think for PV install netboot is the most obvious choice for the
installer, cdrom a distant second and floppy etc nowhere insight)
> It's also awfully i386 specific.
The PAE vs non-PAE problem is pretty i386 specific anyway. Do you have
an idea for a more generic solution? I'm happy to go down a different
route.
> Also, I would personally prefer not to have the full 'uname -r' passed to
> the kernel selection scripts.
yes, this was more of a quick POC hack to solicit opinions on the
general idea -- I'll make it only pass the flavour bit of the version.
> Finally, you seem to be setting a PAE envvar, which is subsequently not used
> anywhere.
Yes, it's a remnant of the early versions of the code. I was half
thinking of adding a check for >4G RAM && PAE for native use too.
Ian.
--
Ian Campbell
Current Noise: Sabbat - The Best Of Enemies
The odds are a million to one against your being one in a million.
Reply to: