Re: [PATCH v3] hw/ppc: Implement -dtb support for PowerNV

[Aditya Gupta <adityag@linux.ibm.com>]

Hello Nick,

On 16/08/24 07:50, Nicholas Piggin wrote:
> <...snip...>
>>> One little nit is MachineState.fdt vs PnvMachineState.fdt
>>> which is now confusing. I would call the new PnvMachineState member
>>> something like fdt_from_dtb, or fdt_override?
>> I agree. this is confusing. machine->fdt could be used instead ?
> Yeah that could be another option. Test pnv.dtb or add a new bool
> to pnv if you need to check whether the fdt has been provided by
> cmdline.

Sure, I will use machine->fdt. Testing pnv.dtb should be good enough to 
check if -dtb was passed I think.


Regarding the conversation about CAS, I don't have idea on it, other 
than the minimum basics. But thanks to you and Cedric, got to know 
somethings.


Thanks,

Aditya Gupta

>>> The other question... Some machines rebuild fdt at init, others at
>>> reset time. As far as I understood, spapr has to rebuild on reset
>>> because C-A-S call can update the fdt so you have to undo that on
>>> reset.
>> C-A-S is a guest OS hcall. reset is called before the guest OS
>> is started.
> Right, but when you reboot it needs to be reverted to initial
> (pre-CAS) fdt.
>
>>> Did powernv just copy that without really needing it, I wonder?
>>> Maybe that rearranged to just do it at init time (e.g., see
>>> hw/riscv/virt.c which is simpler).
>> The machine is aware of user created devices (on the command line)
>> only at reset time.
> Ah, I should have followed a bit closer. riscv, arm use a
> machine_done notifier for that (and x86, loongarch for ACPI / BIOS
> tables). So that avoids fdt rebuild after the first reset I think.
>
> Anyway I don't really mind then, following other archs would be okay,
> but keeping similar with spapr and avoiding code change is also good.
> Maybe add a small comment to we use reset rather than machine_done
> notifier of other archs to be similar to spapr.
>
> Thanks,
> Nick