Re: [RFC v2 2/2] hw/riscv: Add server platform reference machine

[Atish Kumar Patra <atishp@rivosinc.com>]

On Fri, Mar 22, 2024 at 2:20 AM Marcin Juszkiewicz
<marcin.juszkiewicz@linaro.org> wrote:
>
> W dniu 22.03.2024 o 09:50, Heinrich Schuchardt pisze:
>  >>> I see no mention of device trees in the spec, but I do see ACPI. Do we
>  >>> really expect a server platform to use DTs?
>  >>
>  >> This platform "kind of" follows sbsa-ref where we have very
>  >> minimalistic device tree sharing information qemu->firmware.
>  >>
>  >> libfdt is small, format is known and describes hardware. Firmware is
>  >> free to make use of it in any way it wants.
>  >>
>  >> On sbsa-ref we parse DT in TF-A (base firmware) and provide hardware
>  >> information to higher level (edk2) via SMC mechanism. Then EDK2
>  >> creates ACPI tables and provide them to the Operating System.
>
>  > We should ensure that only either an ACPI table or a device-tree
>  > description is passed to the OS and not both, e.g. when using
>  >
>  >      qemu-system-riscv64 -kernel vmlinux -M sbsa-ref
>  >
>  > But that requirement is not machine specific.
>
> I would not call "qemu-system-* -M machinename -k kernel_image" a proper
> way to boot for several systems emulated by QEMU.
>
> DeviceTree is in rvsp-ref and sbsa-ref because it is easy to process in
> limited space 1st stage of firmware has.
>

OpenSBI also has DT support only. So a minimalistic DT generated by the machine
for the firmware is required for RISC-V as well.

> And if we knew how people will mention 'sbsa-ref uses DT' we would use
> something else instead. But that would require adding more code into
> existing firmware projects (libfdt is usually already there).
>
> I did not looked at DT generated for rvsp-ref. I know that sbsa-ref one
> is too minimalistic for kernel use as we added only those fields/nodes
> we need to provide data for firmware.