Re: [PATCH 1/1 RFC] treewide: Deprecate OF_PRIOR_STAGE

[Simon Glass <sjg@chromium.org>]

Hi Thomas,

On Wed, 13 Oct 2021 at 10:22, Thomas Fitzsimmons <fitzsim@fitzsim.org> wrote:
>
> Hi Simon,
>
> Simon Glass <sjg@chromium.org> writes:
>
> > Hi Mark,
> >
> > On Sat, 25 Sept 2021 at 11:27, Mark Kettenis <mark.kettenis@xs4all.nl> wrote:
> >>
> >> > From: Simon Glass <sjg@chromium.org>
> >> > Date: Fri, 24 Sep 2021 07:57:00 -0600
> >> >
> >> > Hi Ilias,
> >> >
> >> > On Fri, 24 Sept 2021 at 07:10, Ilias Apalodimas
> >> > <ilias.apalodimas@linaro.org> wrote:
> >> > >
> >> > > At some point back in 2018 prior_stage_fdt_address and OF_PRIOR_STAGE got
> >> > > introduced,  in order to support a DTB handed over by an earlier stage boot
> >> > > loader.  However we have another option in the Kconfig (OF_BOARD) which has
> >> > > identical semantics.
> >> > >
> >> > > A good example of this is RISC-V boards which during their startup,
> >> > > pick up the DTB from a1 and copy it in their private gd_t.  Apart from that
> >> > > they also copy it to prior_stage_fdt_address,  if the Kconfig option is
> >> > > selected,  which seems unnecessary(??).
> >> > >
> >> > > This is mostly an RFC,  trying to figure out if I am missing some subtle
> >> > > functionality,  which would justify having 2 Kconfig options doing similar
> >> > > things present.
> >> > >
> >> > > - Should we do this?
> >> >
> >> > I think one option is better than two. I have a slight preference for
> >> > OF_PRIOR_STAGE because it is board-agnostic, but I'm not sure it
> >> > matters, since some of these boards are doing strange things anyway
> >> > and cannot use OF_PRIOR_STAGE. So let's go with this.
> >> >
> >> > > - Doesn't OF_BOARD and OF_PRIOR_STAGE practically mean "Someone else is
> >> > >   going to pass me my DTB".  Why should we care if that someone is a prior
> >> > >   bootloader or runtime memory generated on the fly by U-Boot?  It all
> >> > >   boils down to having a *board* specific callback for that.
> >> >
> >> > More generally, I think OF_BOARD is basically 'opt out of the normal
> >> > flow and do something special'.
> >> >
> >> > So at some point I would like to define what 'normal' is. At present,
> >> > normal is OF_SEPARATE which means that the devicetree is packed with
> >> > U-Boot.
> >> >
> >> > Really we want to add a second 'normal', to permit a devicetree (and
> >> > perhaps other stuff) to be passed in. I think this should be that a
> >> > bloblist is passed in, which can contain a devicetree. If it does,
> >> > then the one in U-Boot is ignored.
> >> >
> >> > There should be a standard way to do this with U-Boot. Apart from the
> >> > arch-specific selection of machine registers, the standard way should
> >> > work for all boards, at some indeterminate point in the future.
> >>
> >> There are well-established ABIs here that you can't really change.
> >> One of those ABIs is how the Linux kernel expects to be called.  On
> >> both riscv and arm64 Linux expects to find a pointer to the DTB in a
> >> register.
> >>
> >> Several U-Boot platforms take advantage of this by pretending to be a
> >> Linux kernel.  This way they can be loaded by prior stage firmware
> >> that was designed to directly load a Linux kernel.  This is what I do
> >> for the Apple M1, but this is also how chainloading works on some
> >> chromebooks, and there are a few platforms where a proprietary closed
> >> source first stage bootloader is used.
> >>
> >> So once again, U-Boot should be flexible here.  We can certainly
> >> recommend a certain approach to folks that are building a firmware
> >> stack for new platforms, but we can't really enforce it.
> >
> > Indeed.
> >
> > We can nudge people along by providing useful features. Private
> > firmware does not seem to be going away.
>
> The situation Mark described is the same as the one I was addressing by
> introducing CONFIG_OF_PRIOR_STAGE for these BOLT-using Broadcom boards.
> BOLT is a Broadcom proprietary first- and second-stage bootloader.  On
> these boards, the DTB that BOLT generates in-memory and provides to the
> Linux kernel is meant to be authoritative.
>
> I would much prefer if Broadcom provided native U-Boot support as an
> alternative to BOLT, including maintaining free software device trees.
> But in the absence of that, given that I wanted U-Boot features on these
> boards, I made U-Boot an intermediate (third) stage and used the
> BOLT-provided DTB.  One reason I had for contributing the changes is
> that I was faintly hoping to nudge Broadcom to support these and future
> boards in U-Boot.

There seems to be some genetic quirk in humans that makes them want to
invent their own bootloader. I have seen it a lot in my life.

>
> My understanding is that the DTB design intent does allow things like
> loading a DTB from ROM, so I'm sort of treating the BOLT-provided DTB
> that way.  But I also understand that not having U-Boot and Linux in
> full control of device trees for boards they support is annoying.  That
> said, I'm glad the consensus here seems to be that it's preferable to
> have U-Boot accommodate/still be usable on no-DTS boards.

Yes, that seems clear.

Regards,
Simon