[U-Boot] UBoot running UBoot - is it possible?

[Simon Goldschmidt]

Am Di., 13. Nov. 2018, 17:53 hat Stefano Babic <sbabic at denx.de> geschrieben:

> On 13/11/18 17:07, Simon Goldschmidt wrote:
> > On 13.11.2018 17:00, Wolfgang Denk wrote:
> >> Dear Simon,
> >>
> >> In message
> >> <CAAh8qsxB3YNoDruZnmcPvkygc7HXVOcf=PNdA4xRp=PjmQANsw at mail.gmail.com>
> >> you wrote:
> >>> My idea was to let SPL implement a dedicated boot counter/watchdog
> >>> that detects problems starting U-Boot and using a backup copy if it
> >>> fails multiple times. Of course you need an SPL on your board to do
> >>> this. Plus you end up with the same problem for updating SPL, but I'm
> >>> lucky here that my platform (socfpga gen5) has redundant storage for
> >>> SPL and implements a startup watchdog mechanism for the SPL.
> >> In other words, you are just adding enough additional complexity to
> >> SPL to make it reasonably likely that there will be bugs that need
> >> to be fixed later, i. e. you have to update the SPL.
> >>
> >> And then?
> >>
> >> You are just moving the problem, not solving it.
> >
> > That's not how I see it. As I see it, I have to implement an upgrade
> > option for SPL. This is partly due to bad design of the socfpga_gen5
> > platform. But also I have read multiple times on this list that you
> > should use SPL from the same version as U-Boot as they might work
> > combined and U-Boot might depend on SPL to do things that might change
> > over time.
>
> Right. It is not guaranteed that mixing versions works.
>
> > So is it really a good idea to upgrade U-Boot without
> > upgrading SPL at the same time?
>
> IMHO it is a bad idea, and it forgets that the bootloader is really SPL
> + u-boot.img else just u-boot.img. It is also questionable if it makes
> sense to provide an update mechanism for u-boot when most of critical
> parts like clocks, DDR initialisation, etc. are in SPL.
>
> > It seems to me this would require
> > thorough testing of different version mixes...
> >
>
> Agree, and this becomes a mess.
>
> > So given that SPL must be upgradable, how is it more complex to detect
> > U-Boot failure from SPL than from U-Boot itself?
>
> I do not get the question - if SPL fails, it can be at any time before
> you get the control. There sould be a mechanism to switch to a previous
> copy of SPL, generally this is not available in hw.
>

Unless I'm mistaken, the socfpga gen5 bootrom implements a timer and loads
SPL from secondary storage of it does not register as successful after some
time.

That's my arch only, of course...

Simon