Re: [PATCH v4 03/11] ppc/pnv: New powernv10-rainier machine type

[Nicholas Piggin]

On Tue Nov 21, 2023 at 5:29 PM AEST, Cédric Le Goater wrote:
> On 11/21/23 02:33, Nicholas Piggin wrote:
> > On Tue Nov 21, 2023 at 9:51 AM AEST, Glenn Miles wrote:
> >> Create a new powernv machine type, powernv10-rainier, that
> >> will contain rainier-specific devices.
> > 
> > Is the plan to have a base powernv10 common to all and then
> > powernv10-rainier looks like a Rainier? Or would powernv10
> > just be a rainier?
> > 
> > It's fine to structure code this way, I'm just wondering about
> > the machine types available to user. Is a base powernv10 machine
> > useful to run?
>
> There are multiple P10 boards defined in Linux :
>
>    aspeed-bmc-ibm-bonnell.dts
>    aspeed-bmc-ibm-everest.dts
>    aspeed-bmc-ibm-rainier-1s4u.dts
>    aspeed-bmc-ibm-rainier-4u.dts
>    aspeed-bmc-ibm-rainier.dts
>
> and we could model the machines above with a fixed number of sockets.
> The "powernv10" would be the generic system that can be customized
> at will on the command line, even I2C devices.

If a bare qemu machine could be useful, I don't have a problem with
it. I'm more thinking of what an average OPAL/PowerNV Linux user
developer would want, they (I) would probably want to use powernv,
powernv9, or powernv10, and just get a reasonable "realistic" machine.

The bare system could be powernv10-generic or powernv10-minimal for
those who know what they're doing.

> There is also the
> P10 Denali which is FSP based. This QEMU machine would certainly be
> very different. I thought of doing the same for P9 with a -zaius
> and include NPU2 models for it. I lacked time and the interest was
> small at the time of OpenPOWER.
>
> Anyhow, adding a new machine makes sense and it prepares ground for
> possible new ones. I am OK with or without. As primary users, you are
> the ones that can tell if there will be a second machine.

Yeah we will want to add other machines at some point, I think
this does make sense, my only real concern is what we call them.

Thanks,
Nick