Re: [PATCH 0/3] Add OpenSBI dynamic firmware support

[Alexander Richardson <Alexander.Richardson@cl.cam.ac.uk>]

On Thu, 18 Jun 2020 at 19:22, Atish Patra <atishp@atishpatra.org> wrote:
>
> On Thu, Jun 18, 2020 at 1:56 AM Bin Meng <bmeng.cn@gmail.com> wrote:
> >
> > On Wed, Jun 17, 2020 at 3:29 AM Atish Patra <atish.patra@wdc.com> wrote:
> > >
> > > This series adds support OpenSBI dynamic firmware support to Qemu.
> > > Qemu loader passes the information about the DT and next stage (i.e. kernel
> > > or U-boot) via "a2" register. It allows the user to build bigger OS images
> > > without worrying about overwriting DT. It also unifies the reset vector code
> >
> > I am not sure in what situation overwriting DT could happen. Could you
> > please elaborate?
> >
>
> Currently, the DT is loaded 0x82200000 (34MB offset) for fw_jump.
> Thus, a bigger kernel image
> would overwrite the DT. In fact, it was reported by FreeBSD folks.
> https://github.com/riscv/opensbi/issues/169
>
The problem is that the DT overwrites the kernel image. Usually this
is not noticeable since it's so small and rarely overwrites something
useful, but in my case it was overwriting program memory which
resulted in invalid instruction crashes.
Since this is quite awkward to debug, I added a kernel assertion to
FreeBSD to abort boot in that case.

> There are temporary solutions that can put DT a little bit further or
> put it within 2MB offset. But that's
> just delaying the inevitable.
>
I've changed OpenSBI locally to use a 1MB offset (i.e. place the DT
between OpenSBI and the kernel), but I think the fw_dynamic approach
is much nicer.

Thanks,
Alex