Re: [RFC PATCH qemu] spapr: Kill SLOF

[David Gibson]

On Mon, Jan 13, 2020 at 10:13:27AM +0100, Andrea Bolognani wrote:
> On Wed, 2020-01-08 at 13:34 +1100, Alexey Kardashevskiy wrote:
> > On 07/01/2020 20:39, Andrea Bolognani wrote:
> > > On Tue, 2020-01-07 at 12:55 +1100, Alexey Kardashevskiy wrote:
> > > > Petitboot kernel+initramdisk almost replaces SLOF + GRUB.
> > > 
> > > Is this necessarily a good thing?
> > 
> > The bare metal host and the powernv machine in QEMU do not use grub,
> > they use petitboot which parses all grub configs and supports quite a lot.
> 
> How well does the distro integration work? Eg. if I change something
> in /etc/default/grub and then run grub2-mkconfig, can I expect my
> changes to be picked up?

Yes, in the most trivial possible way.  AIUI, grub needs to process
some of the config file information into a form that the early boot
stages can read before they can read a filesystem.  Petitboot has a
full Linux kernel beneath it so it parses everything from the grub
file at boot time, so if it's in the file it will pick it up.

> In which scenarios will that *not* work?

If you use some grub feature that petitboot doesn't understand /
support.  I don't know off hand of such a feature, but there might be
some.

> > Using Linux for a boot loader is not powerpc-only thing really, some
> > folks do this too (forgot who, just heard this at the KVM forum).
> 
> While other options are available and some architectures use
> something else entirely, GRUB is the de-facto standard across most
> of the non-obscure architectures.
> 
> I guess the question is whether it's more important to be consistent
> within the architecture or across them. I think the latter might be
> preferable, especially when we consider what I think is the most
> common scenario, that is, someone who's used to having GRUB on their
> x86 machine running a ppc64 guest on the cloud. The more skills they
> can automatically transfer over, the better.

Note that killing SLOF isn't necessarily incompatible with running
(true) grub.  It is harder though, because we'd have to fake
significantly more of the OF client interface in qemu to let grub load
the kernel.

> > > Personally I quite like the fact
> > > that I can use the same bootloader across x86, ppc64 and aarch64.
> > 
> > I am not suggesting removing SLOF soon
> 
> Perhaps the patch subject shouldn't be "kill SLOF" then? ;)
> 

-- 
David Gibson                    | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au  | minimalist, thank you.  NOT _the_ _other_
                                | _way_ _around_!
http://www.ozlabs.org/~dgibson

signature.asc
Description: PGP signature